Add files using upload-large-folder tool

.gitattributes

@@ -42,3 +42,7 @@ cudnn-windows-x86_64-8.9.5.30_cuda11-archive/lib/x64/cudnn_ops_infer.lib filter=
 cudnn-windows-x86_64-8.9.5.30_cuda11-archive/lib/x64/cudnn_ops_infer64_8.lib filter=lfs diff=lfs merge=lfs -text
 cudnn-windows-x86_64-8.9.5.30_cuda11-archive/bin/cudnn64_8.dll filter=lfs diff=lfs merge=lfs -text
 pythonProject/.venv/Lib/site-packages/accelerate/utils/__pycache__/dataclasses.cpython-310.pyc filter=lfs diff=lfs merge=lfs -text
+cudnn-windows-x86_64-8.9.5.30_cuda11-archive/lib/x64/cudnn_cnn_infer.lib filter=lfs diff=lfs merge=lfs -text
+cudnn-windows-x86_64-8.9.5.30_cuda11-archive/lib/x64/cudnn_cnn_infer64_8.lib filter=lfs diff=lfs merge=lfs -text
+cudnn-windows-x86_64-8.9.5.30_cuda11-archive/lib/x64/cudnn_cnn_train.lib filter=lfs diff=lfs merge=lfs -text
+cudnn-windows-x86_64-8.9.5.30_cuda11-archive/lib/x64/cudnn_cnn_train64_8.lib filter=lfs diff=lfs merge=lfs -text

cudnn-windows-x86_64-8.9.5.30_cuda11-archive/lib/x64/cudnn_cnn_infer.lib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:957fb42d0fc58c5d043209617ce0ae06ce609d12e8bbab31e3586e93c626f520
+size 2852438

cudnn-windows-x86_64-8.9.5.30_cuda11-archive/lib/x64/cudnn_cnn_infer64_8.lib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:957fb42d0fc58c5d043209617ce0ae06ce609d12e8bbab31e3586e93c626f520
+size 2852438

cudnn-windows-x86_64-8.9.5.30_cuda11-archive/lib/x64/cudnn_cnn_train.lib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69aad4a93f31f68aa34519159b792642150a8a775505a1ed3002110cd03de0fb
+size 1217752

cudnn-windows-x86_64-8.9.5.30_cuda11-archive/lib/x64/cudnn_cnn_train64_8.lib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69aad4a93f31f68aa34519159b792642150a8a775505a1ed3002110cd03de0fb
+size 1217752

pythonProject/.venv/Lib/site-packages/charset_normalizer-3.4.3.dist-info/INSTALLER

@@ -0,0 +1 @@
+pip

pythonProject/.venv/Lib/site-packages/charset_normalizer-3.4.3.dist-info/METADATA

@@ -0,0 +1,750 @@
+Metadata-Version: 2.4
+Name: charset-normalizer
+Version: 3.4.3
+Summary: The Real First Universal Charset Detector. Open, modern and actively maintained alternative to Chardet.
+Author-email: "Ahmed R. TAHRI" <tahri.ahmed@proton.me>
+Maintainer-email: "Ahmed R. TAHRI" <tahri.ahmed@proton.me>
+License: MIT
+Project-URL: Changelog, https://github.com/jawah/charset_normalizer/blob/master/CHANGELOG.md
+Project-URL: Documentation, https://charset-normalizer.readthedocs.io/
+Project-URL: Code, https://github.com/jawah/charset_normalizer
+Project-URL: Issue tracker, https://github.com/jawah/charset_normalizer/issues
+Keywords: encoding,charset,charset-detector,detector,normalization,unicode,chardet,detect
+Classifier: Development Status :: 5 - Production/Stable
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.7
+Classifier: Programming Language :: Python :: 3.8
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: Implementation :: CPython
+Classifier: Programming Language :: Python :: Implementation :: PyPy
+Classifier: Topic :: Text Processing :: Linguistic
+Classifier: Topic :: Utilities
+Classifier: Typing :: Typed
+Requires-Python: >=3.7
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: unicode-backport
+Dynamic: license-file
+
+<h1 align="center">Charset Detection, for Everyone 👋</h1>
+
+<p align="center">
+  <sup>The Real First Universal Charset Detector</sup><br>
+  <a href="https://pypi.org/project/charset-normalizer">
+    <img src="https://img.shields.io/pypi/pyversions/charset_normalizer.svg?orange=blue" />
+  </a>
+  <a href="https://pepy.tech/project/charset-normalizer/">
+    <img alt="Download Count Total" src="https://static.pepy.tech/badge/charset-normalizer/month" />
+  </a>
+  <a href="https://bestpractices.coreinfrastructure.org/projects/7297">
+    <img src="https://bestpractices.coreinfrastructure.org/projects/7297/badge">
+  </a>
+</p>
+<p align="center">
+  <sup><i>Featured Packages</i></sup><br>
+  <a href="https://github.com/jawah/niquests">
+   <img alt="Static Badge" src="https://img.shields.io/badge/Niquests-Most_Advanced_HTTP_Client-cyan">
+  </a>
+  <a href="https://github.com/jawah/wassima">
+   <img alt="Static Badge" src="https://img.shields.io/badge/Wassima-Certifi_Replacement-cyan">
+  </a>
+</p>
+<p align="center">
+  <sup><i>In other language (unofficial port - by the community)</i></sup><br>
+  <a href="https://github.com/nickspring/charset-normalizer-rs">
+   <img alt="Static Badge" src="https://img.shields.io/badge/Rust-red">
+  </a>
+</p>
+
+> A library that helps you read text from an unknown charset encoding.<br /> Motivated by `chardet`,
+> I'm trying to resolve the issue by taking a new approach.
+> All IANA character set names for which the Python core library provides codecs are supported.
+
+<p align="center">
+  >>>>> <a href="https://charsetnormalizerweb.ousret.now.sh" target="_blank">👉 Try Me Online Now, Then Adopt Me 👈 </a> <<<<<
+</p>
+
+This project offers you an alternative to **Universal Charset Encoding Detector**, also known as **Chardet**.
+
+| Feature                                          | [Chardet](https://github.com/chardet/chardet) |                                         Charset Normalizer                                         | [cChardet](https://github.com/PyYoshi/cChardet) |
+|--------------------------------------------------|:---------------------------------------------:|:--------------------------------------------------------------------------------------------------:|:-----------------------------------------------:|
+| `Fast`                                           |                       ❌                       |                                                 ✅                                                  |                        ✅                        |
+| `Universal**`                                    |                       ❌                       |                                                 ✅                                                  |                        ❌                        |
+| `Reliable` **without** distinguishable standards |                       ❌                       |                                                 ✅                                                  |                        ✅                        |
+| `Reliable` **with** distinguishable standards    |                       ✅                       |                                                 ✅                                                  |                        ✅                        |
+| `License`                                        |           LGPL-2.1<br>_restrictive_           |                                                MIT                                                 |            MPL-1.1<br>_restrictive_             |
+| `Native Python`                                  |                       ✅                       |                                                 ✅                                                  |                        ❌                        |
+| `Detect spoken language`                         |                       ❌                       |                                                 ✅                                                  |                       N/A                       |
+| `UnicodeDecodeError Safety`                      |                       ❌                       |                                                 ✅                                                  |                        ❌                        |
+| `Whl Size (min)`                                 |                   193.6 kB                    |                                               42 kB                                                |                     ~200 kB                     |
+| `Supported Encoding`                             |                      33                       | 🎉 [99](https://charset-normalizer.readthedocs.io/en/latest/user/support.html#supported-encodings) |                       40                        |
+
+<p align="center">
+<img src="https://i.imgflip.com/373iay.gif" alt="Reading Normalized Text" width="226"/><img src="https://media.tenor.com/images/c0180f70732a18b4965448d33adba3d0/tenor.gif" alt="Cat Reading Text" width="200"/>
+</p>
+
+*\*\* : They are clearly using specific code for a specific encoding even if covering most of used one*<br>
+
+## ⚡ Performance
+
+This package offer better performance than its counterpart Chardet. Here are some numbers.
+
+| Package                                       | Accuracy | Mean per file (ms) | File per sec (est) |
+|-----------------------------------------------|:--------:|:------------------:|:------------------:|
+| [chardet](https://github.com/chardet/chardet) |   86 %   |       63 ms        |    16 file/sec     |
+| charset-normalizer                            | **98 %** |     **10 ms**      |    100 file/sec    |
+
+| Package                                       | 99th percentile | 95th percentile | 50th percentile |
+|-----------------------------------------------|:---------------:|:---------------:|:---------------:|
+| [chardet](https://github.com/chardet/chardet) |     265 ms      |      71 ms      |      7 ms       |
+| charset-normalizer                            |     100 ms      |      50 ms      |      5 ms       |
+
+_updated as of december 2024 using CPython 3.12_
+
+Chardet's performance on larger file (1MB+) are very poor. Expect huge difference on large payload.
+
+> Stats are generated using 400+ files using default parameters. More details on used files, see GHA workflows.
+> And yes, these results might change at any time. The dataset can be updated to include more files.
+> The actual delays heavily depends on your CPU capabilities. The factors should remain the same.
+> Keep in mind that the stats are generous and that Chardet accuracy vs our is measured using Chardet initial capability
+> (e.g. Supported Encoding) Challenge-them if you want.
+
+## ✨ Installation
+
+Using pip:
+
+```sh
+pip install charset-normalizer -U
+```
+
+## 🚀 Basic Usage
+
+### CLI
+This package comes with a CLI.
+
+```
+usage: normalizer [-h] [-v] [-a] [-n] [-m] [-r] [-f] [-t THRESHOLD]
+                  file [file ...]
+
+The Real First Universal Charset Detector. Discover originating encoding used
+on text file. Normalize text to unicode.
+
+positional arguments:
+  files                 File(s) to be analysed
+
+optional arguments:
+  -h, --help            show this help message and exit
+  -v, --verbose         Display complementary information about file if any.
+                        Stdout will contain logs about the detection process.
+  -a, --with-alternative
+                        Output complementary possibilities if any. Top-level
+                        JSON WILL be a list.
+  -n, --normalize       Permit to normalize input file. If not set, program
+                        does not write anything.
+  -m, --minimal         Only output the charset detected to STDOUT. Disabling
+                        JSON output.
+  -r, --replace         Replace file when trying to normalize it instead of
+                        creating a new one.
+  -f, --force           Replace file without asking if you are sure, use this
+                        flag with caution.
+  -t THRESHOLD, --threshold THRESHOLD
+                        Define a custom maximum amount of chaos allowed in
+                        decoded content. 0. <= chaos <= 1.
+  --version             Show version information and exit.
+```
+
+```bash
+normalizer ./data/sample.1.fr.srt
+```
+
+or
+
+```bash
+python -m charset_normalizer ./data/sample.1.fr.srt
+```
+
+🎉 Since version 1.4.0 the CLI produce easily usable stdout result in JSON format.
+
+```json
+{
+    "path": "/home/default/projects/charset_normalizer/data/sample.1.fr.srt",
+    "encoding": "cp1252",
+    "encoding_aliases": [
+        "1252",
+        "windows_1252"
+    ],
+    "alternative_encodings": [
+        "cp1254",
+        "cp1256",
+        "cp1258",
+        "iso8859_14",
+        "iso8859_15",
+        "iso8859_16",
+        "iso8859_3",
+        "iso8859_9",
+        "latin_1",
+        "mbcs"
+    ],
+    "language": "French",
+    "alphabets": [
+        "Basic Latin",
+        "Latin-1 Supplement"
+    ],
+    "has_sig_or_bom": false,
+    "chaos": 0.149,
+    "coherence": 97.152,
+    "unicode_path": null,
+    "is_preferred": true
+}
+```
+
+### Python
+*Just print out normalized text*
+```python
+from charset_normalizer import from_path
+
+results = from_path('./my_subtitle.srt')
+
+print(str(results.best()))
+```
+
+*Upgrade your code without effort*
+```python
+from charset_normalizer import detect
+```
+
+The above code will behave the same as **chardet**. We ensure that we offer the best (reasonable) BC result possible.
+
+See the docs for advanced usage : [readthedocs.io](https://charset-normalizer.readthedocs.io/en/latest/)
+
+## 😇 Why
+
+When I started using Chardet, I noticed that it was not suited to my expectations, and I wanted to propose a
+reliable alternative using a completely different method. Also! I never back down on a good challenge!
+
+I **don't care** about the **originating charset** encoding, because **two different tables** can
+produce **two identical rendered string.**
+What I want is to get readable text, the best I can.
+
+In a way, **I'm brute forcing text decoding.** How cool is that ? 😎
+
+Don't confuse package **ftfy** with charset-normalizer or chardet. ftfy goal is to repair Unicode string whereas charset-normalizer to convert raw file in unknown encoding to unicode.
+
+## 🍰 How
+
+  - Discard all charset encoding table that could not fit the binary content.
+  - Measure noise, or the mess once opened (by chunks) with a corresponding charset encoding.
+  - Extract matches with the lowest mess detected.
+  - Additionally, we measure coherence / probe for a language.
+
+**Wait a minute**, what is noise/mess and coherence according to **YOU ?**
+
+*Noise :* I opened hundred of text files, **written by humans**, with the wrong encoding table. **I observed**, then
+**I established** some ground rules about **what is obvious** when **it seems like** a mess (aka. defining noise in rendered text).
+ I know that my interpretation of what is noise is probably incomplete, feel free to contribute in order to
+ improve or rewrite it.
+
+*Coherence :* For each language there is on earth, we have computed ranked letter appearance occurrences (the best we can). So I thought
+that intel is worth something here. So I use those records against decoded text to check if I can detect intelligent design.
+
+## ⚡ Known limitations
+
+  - Language detection is unreliable when text contains two or more languages sharing identical letters. (eg. HTML (english tags) + Turkish content (Sharing Latin characters))
+  - Every charset detector heavily depends on sufficient content. In common cases, do not bother run detection on very tiny content.
+
+## ⚠️ About Python EOLs
+
+**If you are running:**
+
+- Python >=2.7,<3.5: Unsupported
+- Python 3.5: charset-normalizer < 2.1
+- Python 3.6: charset-normalizer < 3.1
+- Python 3.7: charset-normalizer < 4.0
+
+Upgrade your Python interpreter as soon as possible.
+
+## 👤 Contributing
+
+Contributions, issues and feature requests are very much welcome.<br />
+Feel free to check [issues page](https://github.com/ousret/charset_normalizer/issues) if you want to contribute.
+
+## 📝 License
+
+Copyright © [Ahmed TAHRI @Ousret](https://github.com/Ousret).<br />
+This project is [MIT](https://github.com/Ousret/charset_normalizer/blob/master/LICENSE) licensed.
+
+Characters frequencies used in this project © 2012 [Denny Vrandečić](http://simia.net/letters/)
+
+## 💼 For Enterprise
+
+Professional support for charset-normalizer is available as part of the [Tidelift
+Subscription][1]. Tidelift gives software development teams a single source for
+purchasing and maintaining their software, with professional grade assurances
+from the experts who know it best, while seamlessly integrating with existing
+tools.
+
+[1]: https://tidelift.com/subscription/pkg/pypi-charset-normalizer?utm_source=pypi-charset-normalizer&utm_medium=readme
+
+[![OpenSSF Best Practices](https://www.bestpractices.dev/projects/7297/badge)](https://www.bestpractices.dev/projects/7297)
+
+# Changelog
+All notable changes to charset-normalizer will be documented in this file. This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
+
+## [3.4.3](https://github.com/Ousret/charset_normalizer/compare/3.4.2...3.4.3) (2025-08-09)
+
+### Changed
+- mypy(c) is no longer a required dependency at build time if `CHARSET_NORMALIZER_USE_MYPYC` isn't set to `1`. (#595) (#583)
+- automatically lower confidence on small bytes samples that are not Unicode in `detect` output legacy function. (#391)
+
+### Added
+- Custom build backend to overcome inability to mark mypy as an optional dependency in the build phase.
+- Support for Python 3.14
+
+### Fixed
+- sdist archive contained useless directories.
+- automatically fallback on valid UTF-16 or UTF-32 even if the md says it's noisy. (#633)
+
+### Misc
+- SBOM are automatically published to the relevant GitHub release to comply with regulatory changes.
+  Each published wheel comes with its SBOM. We choose CycloneDX as the format.
+- Prebuilt optimized wheel are no longer distributed by default for CPython 3.7 due to a change in cibuildwheel.
+
+## [3.4.2](https://github.com/Ousret/charset_normalizer/compare/3.4.1...3.4.2) (2025-05-02)
+
+### Fixed
+- Addressed the DeprecationWarning in our CLI regarding `argparse.FileType` by backporting the target class into the package. (#591)
+- Improved the overall reliability of the detector with CJK Ideographs. (#605) (#587)
+
+### Changed
+- Optional mypyc compilation upgraded to version 1.15 for Python >= 3.8
+
+## [3.4.1](https://github.com/Ousret/charset_normalizer/compare/3.4.0...3.4.1) (2024-12-24)
+
+### Changed
+- Project metadata are now stored using `pyproject.toml` instead of `setup.cfg` using setuptools as the build backend.
+- Enforce annotation delayed loading for a simpler and consistent types in the project.
+- Optional mypyc compilation upgraded to version 1.14 for Python >= 3.8
+
+### Added
+- pre-commit configuration.
+- noxfile.
+
+### Removed
+- `build-requirements.txt` as per using `pyproject.toml` native build configuration.
+- `bin/integration.py` and `bin/serve.py` in favor of downstream integration test (see noxfile).
+- `setup.cfg` in favor of `pyproject.toml` metadata configuration.
+- Unused `utils.range_scan` function.
+
+### Fixed
+- Converting content to Unicode bytes may insert `utf_8` instead of preferred `utf-8`. (#572)
+- Deprecation warning "'count' is passed as positional argument" when converting to Unicode bytes on Python 3.13+
+
+## [3.4.0](https://github.com/Ousret/charset_normalizer/compare/3.3.2...3.4.0) (2024-10-08)
+
+### Added
+- Argument `--no-preemptive` in the CLI to prevent the detector to search for hints.
+- Support for Python 3.13 (#512)
+
+### Fixed
+- Relax the TypeError exception thrown when trying to compare a CharsetMatch with anything else than a CharsetMatch.
+- Improved the general reliability of the detector based on user feedbacks. (#520) (#509) (#498) (#407) (#537)
+- Declared charset in content (preemptive detection) not changed when converting to utf-8 bytes. (#381)
+
+## [3.3.2](https://github.com/Ousret/charset_normalizer/compare/3.3.1...3.3.2) (2023-10-31)
+
+### Fixed
+- Unintentional memory usage regression when using large payload that match several encoding (#376)
+- Regression on some detection case showcased in the documentation (#371)
+
+### Added
+- Noise (md) probe that identify malformed arabic representation due to the presence of letters in isolated form (credit to my wife)
+
+## [3.3.1](https://github.com/Ousret/charset_normalizer/compare/3.3.0...3.3.1) (2023-10-22)
+
+### Changed
+- Optional mypyc compilation upgraded to version 1.6.1 for Python >= 3.8
+- Improved the general detection reliability based on reports from the community
+
+## [3.3.0](https://github.com/Ousret/charset_normalizer/compare/3.2.0...3.3.0) (2023-09-30)
+
+### Added
+- Allow to execute the CLI (e.g. normalizer) through `python -m charset_normalizer.cli` or `python -m charset_normalizer`
+- Support for 9 forgotten encoding that are supported by Python but unlisted in `encoding.aliases` as they have no alias (#323)
+
+### Removed
+- (internal) Redundant utils.is_ascii function and unused function is_private_use_only
+- (internal) charset_normalizer.assets is moved inside charset_normalizer.constant
+
+### Changed
+- (internal) Unicode code blocks in constants are updated using the latest v15.0.0 definition to improve detection
+- Optional mypyc compilation upgraded to version 1.5.1 for Python >= 3.8
+
+### Fixed
+- Unable to properly sort CharsetMatch when both chaos/noise and coherence were close due to an unreachable condition in \_\_lt\_\_ (#350)
+
+## [3.2.0](https://github.com/Ousret/charset_normalizer/compare/3.1.0...3.2.0) (2023-06-07)
+
+### Changed
+- Typehint for function `from_path` no longer enforce `PathLike` as its first argument
+- Minor improvement over the global detection reliability
+
+### Added
+- Introduce function `is_binary` that relies on main capabilities, and optimized to detect binaries
+- Propagate `enable_fallback` argument throughout `from_bytes`, `from_path`, and `from_fp` that allow a deeper control over the detection (default True)
+- Explicit support for Python 3.12
+
+### Fixed
+- Edge case detection failure where a file would contain 'very-long' camel cased word (Issue #289)
+
+## [3.1.0](https://github.com/Ousret/charset_normalizer/compare/3.0.1...3.1.0) (2023-03-06)
+
+### Added
+- Argument `should_rename_legacy` for legacy function `detect` and disregard any new arguments without errors (PR #262)
+
+### Removed
+- Support for Python 3.6 (PR #260)
+
+### Changed
+- Optional speedup provided by mypy/c 1.0.1
+
+## [3.0.1](https://github.com/Ousret/charset_normalizer/compare/3.0.0...3.0.1) (2022-11-18)
+
+### Fixed
+- Multi-bytes cutter/chunk generator did not always cut correctly (PR #233)
+
+### Changed
+- Speedup provided by mypy/c 0.990 on Python >= 3.7
+
+## [3.0.0](https://github.com/Ousret/charset_normalizer/compare/2.1.1...3.0.0) (2022-10-20)
+
+### Added
+- Extend the capability of explain=True when cp_isolation contains at most two entries (min one), will log in details of the Mess-detector results
+- Support for alternative language frequency set in charset_normalizer.assets.FREQUENCIES
+- Add parameter `language_threshold` in `from_bytes`, `from_path` and `from_fp` to adjust the minimum expected coherence ratio
+- `normalizer --version` now specify if current version provide extra speedup (meaning mypyc compilation whl)
+
+### Changed
+- Build with static metadata using 'build' frontend
+- Make the language detection stricter
+- Optional: Module `md.py` can be compiled using Mypyc to provide an extra speedup up to 4x faster than v2.1
+
+### Fixed
+- CLI with opt --normalize fail when using full path for files
+- TooManyAccentuatedPlugin induce false positive on the mess detection when too few alpha character have been fed to it
+- Sphinx warnings when generating the documentation
+
+### Removed
+- Coherence detector no longer return 'Simple English' instead return 'English'
+- Coherence detector no longer return 'Classical Chinese' instead return 'Chinese'
+- Breaking: Method `first()` and `best()` from CharsetMatch
+- UTF-7 will no longer appear as "detected" without a recognized SIG/mark (is unreliable/conflict with ASCII)
+- Breaking: Class aliases CharsetDetector, CharsetDoctor, CharsetNormalizerMatch and CharsetNormalizerMatches
+- Breaking: Top-level function `normalize`
+- Breaking: Properties `chaos_secondary_pass`, `coherence_non_latin` and `w_counter` from CharsetMatch
+- Support for the backport `unicodedata2`
+
+## [3.0.0rc1](https://github.com/Ousret/charset_normalizer/compare/3.0.0b2...3.0.0rc1) (2022-10-18)
+
+### Added
+- Extend the capability of explain=True when cp_isolation contains at most two entries (min one), will log in details of the Mess-detector results
+- Support for alternative language frequency set in charset_normalizer.assets.FREQUENCIES
+- Add parameter `language_threshold` in `from_bytes`, `from_path` and `from_fp` to adjust the minimum expected coherence ratio
+
+### Changed
+- Build with static metadata using 'build' frontend
+- Make the language detection stricter
+
+### Fixed
+- CLI with opt --normalize fail when using full path for files
+- TooManyAccentuatedPlugin induce false positive on the mess detection when too few alpha character have been fed to it
+
+### Removed
+- Coherence detector no longer return 'Simple English' instead return 'English'
+- Coherence detector no longer return 'Classical Chinese' instead return 'Chinese'
+
+## [3.0.0b2](https://github.com/Ousret/charset_normalizer/compare/3.0.0b1...3.0.0b2) (2022-08-21)
+
+### Added
+- `normalizer --version` now specify if current version provide extra speedup (meaning mypyc compilation whl)
+
+### Removed
+- Breaking: Method `first()` and `best()` from CharsetMatch
+- UTF-7 will no longer appear as "detected" without a recognized SIG/mark (is unreliable/conflict with ASCII)
+
+### Fixed
+- Sphinx warnings when generating the documentation
+
+## [3.0.0b1](https://github.com/Ousret/charset_normalizer/compare/2.1.0...3.0.0b1) (2022-08-15)
+
+### Changed
+- Optional: Module `md.py` can be compiled using Mypyc to provide an extra speedup up to 4x faster than v2.1
+
+### Removed
+- Breaking: Class aliases CharsetDetector, CharsetDoctor, CharsetNormalizerMatch and CharsetNormalizerMatches
+- Breaking: Top-level function `normalize`
+- Breaking: Properties `chaos_secondary_pass`, `coherence_non_latin` and `w_counter` from CharsetMatch
+- Support for the backport `unicodedata2`
+
+## [2.1.1](https://github.com/Ousret/charset_normalizer/compare/2.1.0...2.1.1) (2022-08-19)
+
+### Deprecated
+- Function `normalize` scheduled for removal in 3.0
+
+### Changed
+- Removed useless call to decode in fn is_unprintable (#206)
+
+### Fixed
+- Third-party library (i18n xgettext) crashing not recognizing utf_8 (PEP 263) with underscore from [@aleksandernovikov](https://github.com/aleksandernovikov) (#204)
+
+## [2.1.0](https://github.com/Ousret/charset_normalizer/compare/2.0.12...2.1.0) (2022-06-19)
+
+### Added
+- Output the Unicode table version when running the CLI with `--version` (PR #194)
+
+### Changed
+- Re-use decoded buffer for single byte character sets from [@nijel](https://github.com/nijel) (PR #175)
+- Fixing some performance bottlenecks from [@deedy5](https://github.com/deedy5) (PR #183)
+
+### Fixed
+- Workaround potential bug in cpython with Zero Width No-Break Space located in Arabic Presentation Forms-B, Unicode 1.1 not acknowledged as space (PR #175)
+- CLI default threshold aligned with the API threshold from [@oleksandr-kuzmenko](https://github.com/oleksandr-kuzmenko) (PR #181)
+
+### Removed
+- Support for Python 3.5 (PR #192)
+
+### Deprecated
+- Use of backport unicodedata from `unicodedata2` as Python is quickly catching up, scheduled for removal in 3.0 (PR #194)
+
+## [2.0.12](https://github.com/Ousret/charset_normalizer/compare/2.0.11...2.0.12) (2022-02-12)
+
+### Fixed
+- ASCII miss-detection on rare cases (PR #170)
+
+## [2.0.11](https://github.com/Ousret/charset_normalizer/compare/2.0.10...2.0.11) (2022-01-30)
+
+### Added
+- Explicit support for Python 3.11 (PR #164)
+
+### Changed
+- The logging behavior have been completely reviewed, now using only TRACE and DEBUG levels (PR #163 #165)
+
+## [2.0.10](https://github.com/Ousret/charset_normalizer/compare/2.0.9...2.0.10) (2022-01-04)
+
+### Fixed
+- Fallback match entries might lead to UnicodeDecodeError for large bytes sequence (PR #154)
+
+### Changed
+- Skipping the language-detection (CD) on ASCII (PR #155)
+
+## [2.0.9](https://github.com/Ousret/charset_normalizer/compare/2.0.8...2.0.9) (2021-12-03)
+
+### Changed
+- Moderating the logging impact (since 2.0.8) for specific environments (PR #147)
+
+### Fixed
+- Wrong logging level applied when setting kwarg `explain` to True (PR #146)
+
+## [2.0.8](https://github.com/Ousret/charset_normalizer/compare/2.0.7...2.0.8) (2021-11-24)
+### Changed
+- Improvement over Vietnamese detection (PR #126)
+- MD improvement on trailing data and long foreign (non-pure latin) data (PR #124)
+- Efficiency improvements in cd/alphabet_languages from [@adbar](https://github.com/adbar) (PR #122)
+- call sum() without an intermediary list following PEP 289 recommendations from [@adbar](https://github.com/adbar) (PR #129)
+- Code style as refactored by Sourcery-AI (PR #131)
+- Minor adjustment on the MD around european words (PR #133)
+- Remove and replace SRTs from assets / tests (PR #139)
+- Initialize the library logger with a `NullHandler` by default from [@nmaynes](https://github.com/nmaynes) (PR #135)
+- Setting kwarg `explain` to True will add provisionally (bounded to function lifespan) a specific stream handler (PR #135)
+
+### Fixed
+- Fix large (misleading) sequence giving UnicodeDecodeError (PR #137)
+- Avoid using too insignificant chunk (PR #137)
+
+### Added
+- Add and expose function `set_logging_handler` to configure a specific StreamHandler from [@nmaynes](https://github.com/nmaynes) (PR #135)
+- Add `CHANGELOG.md` entries, format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) (PR #141)
+
+## [2.0.7](https://github.com/Ousret/charset_normalizer/compare/2.0.6...2.0.7) (2021-10-11)
+### Added
+- Add support for Kazakh (Cyrillic) language detection (PR #109)
+
+### Changed
+- Further, improve inferring the language from a given single-byte code page (PR #112)
+- Vainly trying to leverage PEP263 when PEP3120 is not supported (PR #116)
+- Refactoring for potential performance improvements in loops from [@adbar](https://github.com/adbar) (PR #113)
+- Various detection improvement (MD+CD) (PR #117)
+
+### Removed
+- Remove redundant logging entry about detected language(s) (PR #115)
+
+### Fixed
+- Fix a minor inconsistency between Python 3.5 and other versions regarding language detection (PR #117 #102)
+
+## [2.0.6](https://github.com/Ousret/charset_normalizer/compare/2.0.5...2.0.6) (2021-09-18)
+### Fixed
+- Unforeseen regression with the loss of the backward-compatibility with some older minor of Python 3.5.x (PR #100)
+- Fix CLI crash when using --minimal output in certain cases (PR #103)
+
+### Changed
+- Minor improvement to the detection efficiency (less than 1%) (PR #106 #101)
+
+## [2.0.5](https://github.com/Ousret/charset_normalizer/compare/2.0.4...2.0.5) (2021-09-14)
+### Changed
+- The project now comply with: flake8, mypy, isort and black to ensure a better overall quality (PR #81)
+- The BC-support with v1.x was improved, the old staticmethods are restored (PR #82)
+- The Unicode detection is slightly improved (PR #93)
+- Add syntax sugar \_\_bool\_\_ for results CharsetMatches list-container (PR #91)
+
+### Removed
+- The project no longer raise warning on tiny content given for detection, will be simply logged as warning instead (PR #92)
+
+### Fixed
+- In some rare case, the chunks extractor could cut in the middle of a multi-byte character and could mislead the mess detection (PR #95)
+- Some rare 'space' characters could trip up the UnprintablePlugin/Mess detection (PR #96)
+- The MANIFEST.in was not exhaustive (PR #78)
+
+## [2.0.4](https://github.com/Ousret/charset_normalizer/compare/2.0.3...2.0.4) (2021-07-30)
+### Fixed
+- The CLI no longer raise an unexpected exception when no encoding has been found (PR #70)
+- Fix accessing the 'alphabets' property when the payload contains surrogate characters (PR #68)
+- The logger could mislead (explain=True) on detected languages and the impact of one MBCS match (PR #72)
+- Submatch factoring could be wrong in rare edge cases (PR #72)
+- Multiple files given to the CLI were ignored when publishing results to STDOUT. (After the first path) (PR #72)
+- Fix line endings from CRLF to LF for certain project files (PR #67)
+
+### Changed
+- Adjust the MD to lower the sensitivity, thus improving the global detection reliability (PR #69 #76)
+- Allow fallback on specified encoding if any (PR #71)
+
+## [2.0.3](https://github.com/Ousret/charset_normalizer/compare/2.0.2...2.0.3) (2021-07-16)
+### Changed
+- Part of the detection mechanism has been improved to be less sensitive, resulting in more accurate detection results. Especially ASCII. (PR #63)
+- According to the community wishes, the detection will fall back on ASCII or UTF-8 in a last-resort case. (PR #64)
+
+## [2.0.2](https://github.com/Ousret/charset_normalizer/compare/2.0.1...2.0.2) (2021-07-15)
+### Fixed
+- Empty/Too small JSON payload miss-detection fixed. Report from [@tseaver](https://github.com/tseaver) (PR #59)
+
+### Changed
+- Don't inject unicodedata2 into sys.modules from [@akx](https://github.com/akx) (PR #57)
+
+## [2.0.1](https://github.com/Ousret/charset_normalizer/compare/2.0.0...2.0.1) (2021-07-13)
+### Fixed
+- Make it work where there isn't a filesystem available, dropping assets frequencies.json. Report from [@sethmlarson](https://github.com/sethmlarson). (PR #55)
+- Using explain=False permanently disable the verbose output in the current runtime (PR #47)
+- One log entry (language target preemptive) was not show in logs when using explain=True (PR #47)
+- Fix undesired exception (ValueError) on getitem of instance CharsetMatches (PR #52)
+
+### Changed
+- Public function normalize default args values were not aligned with from_bytes (PR #53)
+
+### Added
+- You may now use charset aliases in cp_isolation and cp_exclusion arguments (PR #47)
+
+## [2.0.0](https://github.com/Ousret/charset_normalizer/compare/1.4.1...2.0.0) (2021-07-02)
+### Changed
+- 4x to 5 times faster than the previous 1.4.0 release. At least 2x faster than Chardet.
+- Accent has been made on UTF-8 detection, should perform rather instantaneous.
+- The backward compatibility with Chardet has been greatly improved. The legacy detect function returns an identical charset name whenever possible.
+- The detection mechanism has been slightly improved, now Turkish content is detected correctly (most of the time)
+- The program has been rewritten to ease the readability and maintainability. (+Using static typing)+
+- utf_7 detection has been reinstated.
+
+### Removed
+- This package no longer require anything when used with Python 3.5 (Dropped cached_property)
+- Removed support for these languages: Catalan, Esperanto, Kazakh, Baque, Volapük, Azeri, Galician, Nynorsk, Macedonian, and Serbocroatian.
+- The exception hook on UnicodeDecodeError has been removed.
+
+### Deprecated
+- Methods coherence_non_latin, w_counter, chaos_secondary_pass of the class CharsetMatch are now deprecated and scheduled for removal in v3.0
+
+### Fixed
+- The CLI output used the relative path of the file(s). Should be absolute.
+
+## [1.4.1](https://github.com/Ousret/charset_normalizer/compare/1.4.0...1.4.1) (2021-05-28)
+### Fixed
+- Logger configuration/usage no longer conflict with others (PR #44)
+
+## [1.4.0](https://github.com/Ousret/charset_normalizer/compare/1.3.9...1.4.0) (2021-05-21)
+### Removed
+- Using standard logging instead of using the package loguru.
+- Dropping nose test framework in favor of the maintained pytest.
+- Choose to not use dragonmapper package to help with gibberish Chinese/CJK text.
+- Require cached_property only for Python 3.5 due to constraint. Dropping for every other interpreter version.
+- Stop support for UTF-7 that does not contain a SIG.
+- Dropping PrettyTable, replaced with pure JSON output in CLI.
+
+### Fixed
+- BOM marker in a CharsetNormalizerMatch instance could be False in rare cases even if obviously present. Due to the sub-match factoring process.
+- Not searching properly for the BOM when trying utf32/16 parent codec.
+
+### Changed
+- Improving the package final size by compressing frequencies.json.
+- Huge improvement over the larges payload.
+
+### Added
+- CLI now produces JSON consumable output.
+- Return ASCII if given sequences fit. Given reasonable confidence.
+
+## [1.3.9](https://github.com/Ousret/charset_normalizer/compare/1.3.8...1.3.9) (2021-05-13)
+
+### Fixed
+- In some very rare cases, you may end up getting encode/decode errors due to a bad bytes payload (PR #40)
+
+## [1.3.8](https://github.com/Ousret/charset_normalizer/compare/1.3.7...1.3.8) (2021-05-12)
+
+### Fixed
+- Empty given payload for detection may cause an exception if trying to access the `alphabets` property. (PR #39)
+
+## [1.3.7](https://github.com/Ousret/charset_normalizer/compare/1.3.6...1.3.7) (2021-05-12)
+
+### Fixed
+- The legacy detect function should return UTF-8-SIG if sig is present in the payload. (PR #38)
+
+## [1.3.6](https://github.com/Ousret/charset_normalizer/compare/1.3.5...1.3.6) (2021-02-09)
+
+### Changed
+- Amend the previous release to allow prettytable 2.0 (PR #35)
+
+## [1.3.5](https://github.com/Ousret/charset_normalizer/compare/1.3.4...1.3.5) (2021-02-08)
+
+### Fixed
+- Fix error while using the package with a python pre-release interpreter (PR #33)
+
+### Changed
+- Dependencies refactoring, constraints revised.
+
+### Added
+- Add python 3.9 and 3.10 to the supported interpreters
+
+MIT License
+
+Copyright (c) 2025 TAHRI Ahmed R.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

pythonProject/.venv/Lib/site-packages/charset_normalizer-3.4.3.dist-info/RECORD

@@ -0,0 +1,35 @@
+../../Scripts/normalizer.exe,sha256=BsPzI8MuFiVYFkh7wCZGehs-mT7T9f8cK_aAoTNnH6M,108423
+charset_normalizer-3.4.3.dist-info/INSTALLER,sha256=zuuue4knoyJ-UwPPXg8fezS7VCrXJQrAP7zeNuwvFQg,4
+charset_normalizer-3.4.3.dist-info/METADATA,sha256=tqX3UoI-UkqIN99aZsk646yI4NgMbu1MjlKr6BbITG4,37450
+charset_normalizer-3.4.3.dist-info/RECORD,,
+charset_normalizer-3.4.3.dist-info/WHEEL,sha256=KUuBC6lxAbHCKilKua8R9W_TM71_-9Sg5uEP3uDWcoU,101
+charset_normalizer-3.4.3.dist-info/entry_points.txt,sha256=ADSTKrkXZ3hhdOVFi6DcUEHQRS0xfxDIE_pEz4wLIXA,65
+charset_normalizer-3.4.3.dist-info/licenses/LICENSE,sha256=GFd0hdNwTxpHne2OVzwJds_tMV_S_ReYP6mI2kwvcNE,1092
+charset_normalizer-3.4.3.dist-info/top_level.txt,sha256=7ASyzePr8_xuZWJsnqJjIBtyV8vhEo0wBCv1MPRRi3Q,19
+charset_normalizer/__init__.py,sha256=0NT8MHi7SKq3juMqYfOdrkzjisK0L73lneNHH4qaUAs,1638
+charset_normalizer/__main__.py,sha256=2sj_BS6H0sU25C1bMqz9DVwa6kOK9lchSEbSU-_iu7M,115
+charset_normalizer/__pycache__/__init__.cpython-310.pyc,,
+charset_normalizer/__pycache__/__main__.cpython-310.pyc,,
+charset_normalizer/__pycache__/api.cpython-310.pyc,,
+charset_normalizer/__pycache__/cd.cpython-310.pyc,,
+charset_normalizer/__pycache__/constant.cpython-310.pyc,,
+charset_normalizer/__pycache__/legacy.cpython-310.pyc,,
+charset_normalizer/__pycache__/md.cpython-310.pyc,,
+charset_normalizer/__pycache__/models.cpython-310.pyc,,
+charset_normalizer/__pycache__/utils.cpython-310.pyc,,
+charset_normalizer/__pycache__/version.cpython-310.pyc,,
+charset_normalizer/api.py,sha256=ODy4hX78b3ldTl5sViYPU1yzQ5qkclfgSIFE8BtNrTI,23337
+charset_normalizer/cd.py,sha256=uq8nVxRpR6Guc16ACvOWtL8KO3w7vYaCh8hHisuOyTg,12917
+charset_normalizer/cli/__init__.py,sha256=d9MUx-1V_qD3x9igIy4JT4oC5CU0yjulk7QyZWeRFhg,144
+charset_normalizer/cli/__main__.py,sha256=-pdJCyPywouPyFsC8_eTSgTmvh1YEvgjsvy1WZ0XjaA,13027
+charset_normalizer/cli/__pycache__/__init__.cpython-310.pyc,,
+charset_normalizer/cli/__pycache__/__main__.cpython-310.pyc,,
+charset_normalizer/constant.py,sha256=mCJmYzpBU27Ut9kiNWWoBbhhxQ-aRVw3K7LSwoFwBGI,44728
+charset_normalizer/legacy.py,sha256=ui08NlKqAXU3Y7smK-NFJjEgRRQz9ruM7aNCbT0OOrE,2811
+charset_normalizer/md.cp310-win_amd64.pyd,sha256=BQ208ayzKOrtZHPb785b5Hgvw5tc2WszcfHritUOPnw,10752
+charset_normalizer/md.py,sha256=LSuW2hNgXSgF7JGdRapLAHLuj6pABHiP85LTNAYmu7c,20780
+charset_normalizer/md__mypyc.cp310-win_amd64.pyd,sha256=PZHNdte6DpklIoi1GRxQ21vg2eLyv1_q1dx7v_9yui0,125952
+charset_normalizer/models.py,sha256=ZR2PE-fqf6dASZfqdE5Uhkmr0o1MciSdXOjuNqwkmvg,12754
+charset_normalizer/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+charset_normalizer/utils.py,sha256=XtWIQeOuz7cnGebMzyi4Vvi1JtA84QBSIeR9PDzF7pw,12584
+charset_normalizer/version.py,sha256=laniWEeVCCfwRgYLf_rZ2f0qWaNwWTEXQEfUUL_MMvw,123

pythonProject/.venv/Lib/site-packages/charset_normalizer-3.4.3.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (80.9.0)
+Root-Is-Purelib: false
+Tag: cp310-cp310-win_amd64
+

pythonProject/.venv/Lib/site-packages/charset_normalizer-3.4.3.dist-info/licenses/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 TAHRI Ahmed R.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

pythonProject/.venv/Lib/site-packages/charset_normalizer-3.4.3.dist-info/top_level.txt

@@ -0,0 +1 @@
+charset_normalizer

pythonProject/.venv/Lib/site-packages/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_skyreels_image2video.py

@@ -0,0 +1,804 @@
+# Copyright 2025 The HunyuanVideo Team and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+from transformers import CLIPTextModel, CLIPTokenizer, LlamaModel, LlamaTokenizerFast
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput
+from ...loaders import HunyuanVideoLoraLoaderMixin
+from ...models import AutoencoderKLHunyuanVideo, HunyuanVideoTransformer3DModel
+from ...schedulers import FlowMatchEulerDiscreteScheduler
+from ...utils import is_torch_xla_available, logging, replace_example_docstring
+from ...utils.torch_utils import randn_tensor
+from ...video_processor import VideoProcessor
+from ..pipeline_utils import DiffusionPipeline
+from .pipeline_output import HunyuanVideoPipelineOutput
+
+
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```python
+        >>> import torch
+        >>> from diffusers import HunyuanSkyreelsImageToVideoPipeline, HunyuanVideoTransformer3DModel
+        >>> from diffusers.utils import load_image, export_to_video
+
+        >>> model_id = "hunyuanvideo-community/HunyuanVideo"
+        >>> transformer_model_id = "Skywork/SkyReels-V1-Hunyuan-I2V"
+        >>> transformer = HunyuanVideoTransformer3DModel.from_pretrained(
+        ...     transformer_model_id, torch_dtype=torch.bfloat16
+        ... )
+        >>> pipe = HunyuanSkyreelsImageToVideoPipeline.from_pretrained(
+        ...     model_id, transformer=transformer, torch_dtype=torch.float16
+        ... )
+        >>> pipe.vae.enable_tiling()
+        >>> pipe.to("cuda")
+
+        >>> prompt = "An astronaut hatching from an egg, on the surface of the moon, the darkness and depth of space realised in the background. High quality, ultrarealistic detail and breath-taking movie-like camera shot."
+        >>> negative_prompt = "Aerial view, aerial view, overexposed, low quality, deformation, a poor composition, bad hands, bad teeth, bad eyes, bad limbs, distortion"
+        >>> image = load_image(
+        ...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/astronaut.jpg"
+        ... )
+
+        >>> output = pipe(
+        ...     image=image,
+        ...     prompt=prompt,
+        ...     negative_prompt=negative_prompt,
+        ...     num_inference_steps=30,
+        ...     true_cfg_scale=6.0,
+        ...     guidance_scale=1.0,
+        ... ).frames[0]
+        >>> export_to_video(output, "output.mp4", fps=15)
+        ```
+"""
+
+
+DEFAULT_PROMPT_TEMPLATE = {
+    "template": (
+        "<|start_header_id|>system<|end_header_id|>\n\nDescribe the video by detailing the following aspects: "
+        "1. The main content and theme of the video."
+        "2. The color, shape, size, texture, quantity, text, and spatial relationships of the objects."
+        "3. Actions, events, behaviors temporal relationships, physical movement changes of the objects."
+        "4. background environment, light, style and atmosphere."
+        "5. camera angles, movements, and transitions used in the video:<|eot_id|>"
+        "<|start_header_id|>user<|end_header_id|>\n\n{}<|eot_id|>"
+    ),
+    "crop_start": 95,
+}
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
+    **kwargs,
+):
+    r"""
+    Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+    custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+            must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+    encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+    if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+        return encoder_output.latent_dist.sample(generator)
+    elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+        return encoder_output.latent_dist.mode()
+    elif hasattr(encoder_output, "latents"):
+        return encoder_output.latents
+    else:
+        raise AttributeError("Could not access latents of provided encoder_output")
+
+
+class HunyuanSkyreelsImageToVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
+    r"""
+    Pipeline for image-to-video generation using HunyuanVideo.
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+    implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+    Args:
+        text_encoder ([`LlamaModel`]):
+            [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
+        tokenizer (`LlamaTokenizer`):
+            Tokenizer from [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
+        transformer ([`HunyuanVideoTransformer3DModel`]):
+            Conditional Transformer to denoise the encoded image latents.
+        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
+            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
+        vae ([`AutoencoderKLHunyuanVideo`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
+        text_encoder_2 ([`CLIPTextModel`]):
+            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+        tokenizer_2 (`CLIPTokenizer`):
+            Tokenizer of class
+            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
+    """
+
+    model_cpu_offload_seq = "text_encoder->text_encoder_2->transformer->vae"
+    _callback_tensor_inputs = ["latents", "prompt_embeds"]
+
+    def __init__(
+        self,
+        text_encoder: LlamaModel,
+        tokenizer: LlamaTokenizerFast,
+        transformer: HunyuanVideoTransformer3DModel,
+        vae: AutoencoderKLHunyuanVideo,
+        scheduler: FlowMatchEulerDiscreteScheduler,
+        text_encoder_2: CLIPTextModel,
+        tokenizer_2: CLIPTokenizer,
+    ):
+        super().__init__()
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            transformer=transformer,
+            scheduler=scheduler,
+            text_encoder_2=text_encoder_2,
+            tokenizer_2=tokenizer_2,
+        )
+
+        self.vae_scale_factor_temporal = self.vae.temporal_compression_ratio if getattr(self, "vae", None) else 4
+        self.vae_scale_factor_spatial = self.vae.spatial_compression_ratio if getattr(self, "vae", None) else 8
+        self.vae_scaling_factor = self.vae.config.scaling_factor if getattr(self, "vae", None) else 0.476986
+        self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline._get_llama_prompt_embeds
+    def _get_llama_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]],
+        prompt_template: Dict[str, Any],
+        num_videos_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 256,
+        num_hidden_layers_to_skip: int = 2,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt)
+
+        prompt = [prompt_template["template"].format(p) for p in prompt]
+
+        crop_start = prompt_template.get("crop_start", None)
+        if crop_start is None:
+            prompt_template_input = self.tokenizer(
+                prompt_template["template"],
+                padding="max_length",
+                return_tensors="pt",
+                return_length=False,
+                return_overflowing_tokens=False,
+                return_attention_mask=False,
+            )
+            crop_start = prompt_template_input["input_ids"].shape[-1]
+            # Remove <|eot_id|> token and placeholder {}
+            crop_start -= 2
+
+        max_sequence_length += crop_start
+        text_inputs = self.tokenizer(
+            prompt,
+            max_length=max_sequence_length,
+            padding="max_length",
+            truncation=True,
+            return_tensors="pt",
+            return_length=False,
+            return_overflowing_tokens=False,
+            return_attention_mask=True,
+        )
+        text_input_ids = text_inputs.input_ids.to(device=device)
+        prompt_attention_mask = text_inputs.attention_mask.to(device=device)
+
+        prompt_embeds = self.text_encoder(
+            input_ids=text_input_ids,
+            attention_mask=prompt_attention_mask,
+            output_hidden_states=True,
+        ).hidden_states[-(num_hidden_layers_to_skip + 1)]
+        prompt_embeds = prompt_embeds.to(dtype=dtype)
+
+        if crop_start is not None and crop_start > 0:
+            prompt_embeds = prompt_embeds[:, crop_start:]
+            prompt_attention_mask = prompt_attention_mask[:, crop_start:]
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        _, seq_len, _ = prompt_embeds.shape
+        prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1)
+        prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)
+        prompt_attention_mask = prompt_attention_mask.repeat(1, num_videos_per_prompt)
+        prompt_attention_mask = prompt_attention_mask.view(batch_size * num_videos_per_prompt, seq_len)
+
+        return prompt_embeds, prompt_attention_mask
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline._get_clip_prompt_embeds
+    def _get_clip_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]],
+        num_videos_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 77,
+    ) -> torch.Tensor:
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder_2.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt)
+
+        text_inputs = self.tokenizer_2(
+            prompt,
+            padding="max_length",
+            max_length=max_sequence_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+
+        text_input_ids = text_inputs.input_ids
+        untruncated_ids = self.tokenizer_2(prompt, padding="longest", return_tensors="pt").input_ids
+        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+            removed_text = self.tokenizer_2.batch_decode(untruncated_ids[:, max_sequence_length - 1 : -1])
+            logger.warning(
+                "The following part of your input was truncated because CLIP can only handle sequences up to"
+                f" {max_sequence_length} tokens: {removed_text}"
+            )
+
+        prompt_embeds = self.text_encoder_2(text_input_ids.to(device), output_hidden_states=False).pooler_output
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt)
+        prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, -1)
+
+        return prompt_embeds
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline.encode_prompt
+    def encode_prompt(
+        self,
+        prompt: Union[str, List[str]],
+        prompt_2: Union[str, List[str]] = None,
+        prompt_template: Dict[str, Any] = DEFAULT_PROMPT_TEMPLATE,
+        num_videos_per_prompt: int = 1,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_attention_mask: Optional[torch.Tensor] = None,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 256,
+    ):
+        if prompt_embeds is None:
+            prompt_embeds, prompt_attention_mask = self._get_llama_prompt_embeds(
+                prompt,
+                prompt_template,
+                num_videos_per_prompt,
+                device=device,
+                dtype=dtype,
+                max_sequence_length=max_sequence_length,
+            )
+
+        if pooled_prompt_embeds is None:
+            if prompt_2 is None:
+                prompt_2 = prompt
+            pooled_prompt_embeds = self._get_clip_prompt_embeds(
+                prompt,
+                num_videos_per_prompt,
+                device=device,
+                dtype=dtype,
+                max_sequence_length=77,
+            )
+
+        return prompt_embeds, pooled_prompt_embeds, prompt_attention_mask
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline.check_inputs
+    def check_inputs(
+        self,
+        prompt,
+        prompt_2,
+        height,
+        width,
+        prompt_embeds=None,
+        callback_on_step_end_tensor_inputs=None,
+        prompt_template=None,
+    ):
+        if height % 16 != 0 or width % 16 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 16 but are {height} and {width}.")
+
+        if callback_on_step_end_tensor_inputs is not None and not all(
+            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+        ):
+            raise ValueError(
+                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+            )
+
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt_2 is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+        elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+            raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+        if prompt_template is not None:
+            if not isinstance(prompt_template, dict):
+                raise ValueError(f"`prompt_template` has to be of type `dict` but is {type(prompt_template)}")
+            if "template" not in prompt_template:
+                raise ValueError(
+                    f"`prompt_template` has to contain a key `template` but only found {prompt_template.keys()}"
+                )
+
+    def prepare_latents(
+        self,
+        image: torch.Tensor,
+        batch_size: int,
+        num_channels_latents: int = 32,
+        height: int = 544,
+        width: int = 960,
+        num_frames: int = 97,
+        dtype: Optional[torch.dtype] = None,
+        device: Optional[torch.device] = None,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        image = image.unsqueeze(2)  # [B, C, 1, H, W]
+        if isinstance(generator, list):
+            image_latents = [
+                retrieve_latents(self.vae.encode(image[i].unsqueeze(0)), generator[i]) for i in range(batch_size)
+            ]
+        else:
+            image_latents = [retrieve_latents(self.vae.encode(img.unsqueeze(0)), generator) for img in image]
+
+        image_latents = torch.cat(image_latents, dim=0).to(dtype) * self.vae_scaling_factor
+
+        num_latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
+        latent_height, latent_width = height // self.vae_scale_factor_spatial, width // self.vae_scale_factor_spatial
+        shape = (batch_size, num_channels_latents, num_latent_frames, latent_height, latent_width)
+        padding_shape = (batch_size, num_channels_latents, num_latent_frames - 1, latent_height, latent_width)
+
+        latents_padding = torch.zeros(padding_shape, dtype=dtype, device=device)
+        image_latents = torch.cat([image_latents, latents_padding], dim=2)
+
+        if latents is None:
+            latents = randn_tensor(shape, generator=generator, dtype=dtype, device=device)
+        else:
+            latents = latents.to(dtype=dtype, device=device)
+
+        return latents, image_latents
+
+    def enable_vae_slicing(self):
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.vae.enable_slicing()
+
+    def disable_vae_slicing(self):
+        r"""
+        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_slicing()
+
+    def enable_vae_tiling(self):
+        r"""
+        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+        processing larger images.
+        """
+        self.vae.enable_tiling()
+
+    def disable_vae_tiling(self):
+        r"""
+        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_tiling()
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    @property
+    def num_timesteps(self):
+        return self._num_timesteps
+
+    @property
+    def attention_kwargs(self):
+        return self._attention_kwargs
+
+    @property
+    def current_timestep(self):
+        return self._current_timestep
+
+    @property
+    def interrupt(self):
+        return self._interrupt
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        image: PipelineImageInput,
+        prompt: Union[str, List[str]] = None,
+        prompt_2: Union[str, List[str]] = None,
+        negative_prompt: Union[str, List[str]] = None,
+        negative_prompt_2: Union[str, List[str]] = None,
+        height: int = 544,
+        width: int = 960,
+        num_frames: int = 97,
+        num_inference_steps: int = 50,
+        sigmas: List[float] = None,
+        true_cfg_scale: float = 6.0,
+        guidance_scale: float = 1.0,
+        num_videos_per_prompt: Optional[int] = 1,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_attention_mask: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        attention_kwargs: Optional[Dict[str, Any]] = None,
+        callback_on_step_end: Optional[
+            Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+        ] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        prompt_template: Dict[str, Any] = DEFAULT_PROMPT_TEMPLATE,
+        max_sequence_length: int = 256,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+                will be used instead.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
+                not greater than `1`).
+            negative_prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+                `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
+            height (`int`, defaults to `720`):
+                The height in pixels of the generated image.
+            width (`int`, defaults to `1280`):
+                The width in pixels of the generated image.
+            num_frames (`int`, defaults to `129`):
+                The number of frames in the generated video.
+            num_inference_steps (`int`, defaults to `50`):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
+            true_cfg_scale (`float`, *optional*, defaults to 1.0):
+                When > 1.0 and a provided `negative_prompt`, enables true classifier-free guidance.
+            guidance_scale (`float`, defaults to `6.0`):
+                Guidance scale as defined in [Classifier-Free Diffusion
+                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+                the text `prompt`, usually at the expense of lower image quality. Note that the only available
+                HunyuanVideo model is CFG-distilled, which means that traditional guidance between unconditional and
+                conditional latent is not applied.
+            num_videos_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            latents (`torch.Tensor`, *optional*):
+                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor is generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.Tensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+                input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`HunyuanVideoPipelineOutput`] instead of a plain tuple.
+            attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
+            callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+                A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+                each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+                DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+                list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeline class.
+
+        Examples:
+
+        Returns:
+            [`~HunyuanVideoPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`HunyuanVideoPipelineOutput`] is returned, otherwise a `tuple` is returned
+                where the first element is a list with the generated images and the second element is a list of `bool`s
+                indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw) content.
+        """
+
+        if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+            callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            prompt_2,
+            height,
+            width,
+            prompt_embeds,
+            callback_on_step_end_tensor_inputs,
+            prompt_template,
+        )
+
+        has_neg_prompt = negative_prompt is not None or (
+            negative_prompt_embeds is not None and negative_pooled_prompt_embeds is not None
+        )
+        do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
+
+        self._guidance_scale = guidance_scale
+        self._attention_kwargs = attention_kwargs
+        self._current_timestep = None
+        self._interrupt = False
+
+        device = self._execution_device
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        # 3. Encode input prompt
+        transformer_dtype = self.transformer.dtype
+        prompt_embeds, pooled_prompt_embeds, prompt_attention_mask = self.encode_prompt(
+            prompt=prompt,
+            prompt_2=prompt_2,
+            prompt_template=prompt_template,
+            num_videos_per_prompt=num_videos_per_prompt,
+            prompt_embeds=prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            prompt_attention_mask=prompt_attention_mask,
+            device=device,
+            max_sequence_length=max_sequence_length,
+        )
+        prompt_embeds = prompt_embeds.to(transformer_dtype)
+        prompt_attention_mask = prompt_attention_mask.to(transformer_dtype)
+        pooled_prompt_embeds = pooled_prompt_embeds.to(transformer_dtype)
+
+        if do_true_cfg:
+            negative_prompt_embeds, negative_pooled_prompt_embeds, negative_prompt_attention_mask = self.encode_prompt(
+                prompt=negative_prompt,
+                prompt_2=negative_prompt_2,
+                prompt_template=prompt_template,
+                num_videos_per_prompt=num_videos_per_prompt,
+                prompt_embeds=negative_prompt_embeds,
+                pooled_prompt_embeds=negative_pooled_prompt_embeds,
+                prompt_attention_mask=negative_prompt_attention_mask,
+                device=device,
+                max_sequence_length=max_sequence_length,
+            )
+            negative_prompt_embeds = negative_prompt_embeds.to(transformer_dtype)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.to(transformer_dtype)
+            negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.to(transformer_dtype)
+
+        # 4. Prepare timesteps
+        sigmas = np.linspace(1.0, 0.0, num_inference_steps + 1)[:-1] if sigmas is None else sigmas
+        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, sigmas=sigmas)
+
+        # 5. Prepare latent variables
+        vae_dtype = self.vae.dtype
+        image = self.video_processor.preprocess(image, height=height, width=width).to(device, vae_dtype)
+        num_channels_latents = self.transformer.config.in_channels // 2
+        latents, image_latents = self.prepare_latents(
+            image,
+            batch_size * num_videos_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            num_frames,
+            torch.float32,
+            device,
+            generator,
+            latents,
+        )
+        latent_image_input = image_latents.to(transformer_dtype)
+
+        # 6. Prepare guidance condition
+        guidance = torch.tensor([guidance_scale] * latents.shape[0], dtype=transformer_dtype, device=device) * 1000.0
+
+        # 7. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        self._num_timesteps = len(timesteps)
+
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if self.interrupt:
+                    continue
+
+                self._current_timestep = t
+                latent_model_input = latents.to(transformer_dtype)
+                latent_model_input = torch.cat([latent_model_input, latent_image_input], dim=1)
+
+                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+                timestep = t.expand(latents.shape[0]).to(latents.dtype)
+
+                noise_pred = self.transformer(
+                    hidden_states=latent_model_input,
+                    timestep=timestep,
+                    encoder_hidden_states=prompt_embeds,
+                    encoder_attention_mask=prompt_attention_mask,
+                    pooled_projections=pooled_prompt_embeds,
+                    guidance=guidance,
+                    attention_kwargs=attention_kwargs,
+                    return_dict=False,
+                )[0]
+
+                if do_true_cfg:
+                    neg_noise_pred = self.transformer(
+                        hidden_states=latent_model_input,
+                        timestep=timestep,
+                        encoder_hidden_states=negative_prompt_embeds,
+                        encoder_attention_mask=negative_prompt_attention_mask,
+                        pooled_projections=negative_pooled_prompt_embeds,
+                        guidance=guidance,
+                        attention_kwargs=attention_kwargs,
+                        return_dict=False,
+                    )[0]
+                    noise_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
+        self._current_timestep = None
+
+        if not output_type == "latent":
+            latents = latents.to(self.vae.dtype) / self.vae.config.scaling_factor
+            video = self.vae.decode(latents, return_dict=False)[0]
+            video = self.video_processor.postprocess_video(video, output_type=output_type)
+        else:
+            video = latents
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (video,)
+
+        return HunyuanVideoPipelineOutput(frames=video)

pythonProject/.venv/Lib/site-packages/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py

@@ -0,0 +1,755 @@
+# Copyright 2025 The HunyuanVideo Team and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+from transformers import CLIPTextModel, CLIPTokenizer, LlamaModel, LlamaTokenizerFast
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...loaders import HunyuanVideoLoraLoaderMixin
+from ...models import AutoencoderKLHunyuanVideo, HunyuanVideoTransformer3DModel
+from ...schedulers import FlowMatchEulerDiscreteScheduler
+from ...utils import is_torch_xla_available, logging, replace_example_docstring
+from ...utils.torch_utils import randn_tensor
+from ...video_processor import VideoProcessor
+from ..pipeline_utils import DiffusionPipeline
+from .pipeline_output import HunyuanVideoPipelineOutput
+
+
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```python
+        >>> import torch
+        >>> from diffusers import HunyuanVideoPipeline, HunyuanVideoTransformer3DModel
+        >>> from diffusers.utils import export_to_video
+
+        >>> model_id = "hunyuanvideo-community/HunyuanVideo"
+        >>> transformer = HunyuanVideoTransformer3DModel.from_pretrained(
+        ...     model_id, subfolder="transformer", torch_dtype=torch.bfloat16
+        ... )
+        >>> pipe = HunyuanVideoPipeline.from_pretrained(model_id, transformer=transformer, torch_dtype=torch.float16)
+        >>> pipe.vae.enable_tiling()
+        >>> pipe.to("cuda")
+
+        >>> output = pipe(
+        ...     prompt="A cat walks on the grass, realistic",
+        ...     height=320,
+        ...     width=512,
+        ...     num_frames=61,
+        ...     num_inference_steps=30,
+        ... ).frames[0]
+        >>> export_to_video(output, "output.mp4", fps=15)
+        ```
+"""
+
+
+DEFAULT_PROMPT_TEMPLATE = {
+    "template": (
+        "<|start_header_id|>system<|end_header_id|>\n\nDescribe the video by detailing the following aspects: "
+        "1. The main content and theme of the video."
+        "2. The color, shape, size, texture, quantity, text, and spatial relationships of the objects."
+        "3. Actions, events, behaviors temporal relationships, physical movement changes of the objects."
+        "4. background environment, light, style and atmosphere."
+        "5. camera angles, movements, and transitions used in the video:<|eot_id|>"
+        "<|start_header_id|>user<|end_header_id|>\n\n{}<|eot_id|>"
+    ),
+    "crop_start": 95,
+}
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
+    **kwargs,
+):
+    r"""
+    Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+    custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+            must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
+class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
+    r"""
+    Pipeline for text-to-video generation using HunyuanVideo.
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+    implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+    Args:
+        text_encoder ([`LlamaModel`]):
+            [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
+        tokenizer (`LlamaTokenizer`):
+            Tokenizer from [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
+        transformer ([`HunyuanVideoTransformer3DModel`]):
+            Conditional Transformer to denoise the encoded image latents.
+        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
+            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
+        vae ([`AutoencoderKLHunyuanVideo`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
+        text_encoder_2 ([`CLIPTextModel`]):
+            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+        tokenizer_2 (`CLIPTokenizer`):
+            Tokenizer of class
+            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
+    """
+
+    model_cpu_offload_seq = "text_encoder->text_encoder_2->transformer->vae"
+    _callback_tensor_inputs = ["latents", "prompt_embeds"]
+
+    def __init__(
+        self,
+        text_encoder: LlamaModel,
+        tokenizer: LlamaTokenizerFast,
+        transformer: HunyuanVideoTransformer3DModel,
+        vae: AutoencoderKLHunyuanVideo,
+        scheduler: FlowMatchEulerDiscreteScheduler,
+        text_encoder_2: CLIPTextModel,
+        tokenizer_2: CLIPTokenizer,
+    ):
+        super().__init__()
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            transformer=transformer,
+            scheduler=scheduler,
+            text_encoder_2=text_encoder_2,
+            tokenizer_2=tokenizer_2,
+        )
+
+        self.vae_scale_factor_temporal = self.vae.temporal_compression_ratio if getattr(self, "vae", None) else 4
+        self.vae_scale_factor_spatial = self.vae.spatial_compression_ratio if getattr(self, "vae", None) else 8
+        self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
+
+    def _get_llama_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]],
+        prompt_template: Dict[str, Any],
+        num_videos_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 256,
+        num_hidden_layers_to_skip: int = 2,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt)
+
+        prompt = [prompt_template["template"].format(p) for p in prompt]
+
+        crop_start = prompt_template.get("crop_start", None)
+        if crop_start is None:
+            prompt_template_input = self.tokenizer(
+                prompt_template["template"],
+                padding="max_length",
+                return_tensors="pt",
+                return_length=False,
+                return_overflowing_tokens=False,
+                return_attention_mask=False,
+            )
+            crop_start = prompt_template_input["input_ids"].shape[-1]
+            # Remove <|eot_id|> token and placeholder {}
+            crop_start -= 2
+
+        max_sequence_length += crop_start
+        text_inputs = self.tokenizer(
+            prompt,
+            max_length=max_sequence_length,
+            padding="max_length",
+            truncation=True,
+            return_tensors="pt",
+            return_length=False,
+            return_overflowing_tokens=False,
+            return_attention_mask=True,
+        )
+        text_input_ids = text_inputs.input_ids.to(device=device)
+        prompt_attention_mask = text_inputs.attention_mask.to(device=device)
+
+        prompt_embeds = self.text_encoder(
+            input_ids=text_input_ids,
+            attention_mask=prompt_attention_mask,
+            output_hidden_states=True,
+        ).hidden_states[-(num_hidden_layers_to_skip + 1)]
+        prompt_embeds = prompt_embeds.to(dtype=dtype)
+
+        if crop_start is not None and crop_start > 0:
+            prompt_embeds = prompt_embeds[:, crop_start:]
+            prompt_attention_mask = prompt_attention_mask[:, crop_start:]
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        _, seq_len, _ = prompt_embeds.shape
+        prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1)
+        prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)
+        prompt_attention_mask = prompt_attention_mask.repeat(1, num_videos_per_prompt)
+        prompt_attention_mask = prompt_attention_mask.view(batch_size * num_videos_per_prompt, seq_len)
+
+        return prompt_embeds, prompt_attention_mask
+
+    def _get_clip_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]],
+        num_videos_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 77,
+    ) -> torch.Tensor:
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder_2.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt)
+
+        text_inputs = self.tokenizer_2(
+            prompt,
+            padding="max_length",
+            max_length=max_sequence_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+
+        text_input_ids = text_inputs.input_ids
+        untruncated_ids = self.tokenizer_2(prompt, padding="longest", return_tensors="pt").input_ids
+        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+            removed_text = self.tokenizer_2.batch_decode(untruncated_ids[:, max_sequence_length - 1 : -1])
+            logger.warning(
+                "The following part of your input was truncated because CLIP can only handle sequences up to"
+                f" {max_sequence_length} tokens: {removed_text}"
+            )
+
+        prompt_embeds = self.text_encoder_2(text_input_ids.to(device), output_hidden_states=False).pooler_output
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt)
+        prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, -1)
+
+        return prompt_embeds
+
+    def encode_prompt(
+        self,
+        prompt: Union[str, List[str]],
+        prompt_2: Union[str, List[str]] = None,
+        prompt_template: Dict[str, Any] = DEFAULT_PROMPT_TEMPLATE,
+        num_videos_per_prompt: int = 1,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_attention_mask: Optional[torch.Tensor] = None,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 256,
+    ):
+        if prompt_embeds is None:
+            prompt_embeds, prompt_attention_mask = self._get_llama_prompt_embeds(
+                prompt,
+                prompt_template,
+                num_videos_per_prompt,
+                device=device,
+                dtype=dtype,
+                max_sequence_length=max_sequence_length,
+            )
+
+        if pooled_prompt_embeds is None:
+            if prompt_2 is None:
+                prompt_2 = prompt
+            pooled_prompt_embeds = self._get_clip_prompt_embeds(
+                prompt,
+                num_videos_per_prompt,
+                device=device,
+                dtype=dtype,
+                max_sequence_length=77,
+            )
+
+        return prompt_embeds, pooled_prompt_embeds, prompt_attention_mask
+
+    def check_inputs(
+        self,
+        prompt,
+        prompt_2,
+        height,
+        width,
+        prompt_embeds=None,
+        callback_on_step_end_tensor_inputs=None,
+        prompt_template=None,
+    ):
+        if height % 16 != 0 or width % 16 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 16 but are {height} and {width}.")
+
+        if callback_on_step_end_tensor_inputs is not None and not all(
+            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+        ):
+            raise ValueError(
+                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+            )
+
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt_2 is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+        elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+            raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+        if prompt_template is not None:
+            if not isinstance(prompt_template, dict):
+                raise ValueError(f"`prompt_template` has to be of type `dict` but is {type(prompt_template)}")
+            if "template" not in prompt_template:
+                raise ValueError(
+                    f"`prompt_template` has to contain a key `template` but only found {prompt_template.keys()}"
+                )
+
+    def prepare_latents(
+        self,
+        batch_size: int,
+        num_channels_latents: int = 32,
+        height: int = 720,
+        width: int = 1280,
+        num_frames: int = 129,
+        dtype: Optional[torch.dtype] = None,
+        device: Optional[torch.device] = None,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        if latents is not None:
+            return latents.to(device=device, dtype=dtype)
+
+        shape = (
+            batch_size,
+            num_channels_latents,
+            (num_frames - 1) // self.vae_scale_factor_temporal + 1,
+            int(height) // self.vae_scale_factor_spatial,
+            int(width) // self.vae_scale_factor_spatial,
+        )
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+        return latents
+
+    def enable_vae_slicing(self):
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.vae.enable_slicing()
+
+    def disable_vae_slicing(self):
+        r"""
+        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_slicing()
+
+    def enable_vae_tiling(self):
+        r"""
+        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+        processing larger images.
+        """
+        self.vae.enable_tiling()
+
+    def disable_vae_tiling(self):
+        r"""
+        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_tiling()
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    @property
+    def num_timesteps(self):
+        return self._num_timesteps
+
+    @property
+    def attention_kwargs(self):
+        return self._attention_kwargs
+
+    @property
+    def current_timestep(self):
+        return self._current_timestep
+
+    @property
+    def interrupt(self):
+        return self._interrupt
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        prompt_2: Union[str, List[str]] = None,
+        negative_prompt: Union[str, List[str]] = None,
+        negative_prompt_2: Union[str, List[str]] = None,
+        height: int = 720,
+        width: int = 1280,
+        num_frames: int = 129,
+        num_inference_steps: int = 50,
+        sigmas: List[float] = None,
+        true_cfg_scale: float = 1.0,
+        guidance_scale: float = 6.0,
+        num_videos_per_prompt: Optional[int] = 1,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_attention_mask: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        attention_kwargs: Optional[Dict[str, Any]] = None,
+        callback_on_step_end: Optional[
+            Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+        ] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        prompt_template: Dict[str, Any] = DEFAULT_PROMPT_TEMPLATE,
+        max_sequence_length: int = 256,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+                will be used instead.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
+                not greater than `1`).
+            negative_prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+                `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
+            height (`int`, defaults to `720`):
+                The height in pixels of the generated image.
+            width (`int`, defaults to `1280`):
+                The width in pixels of the generated image.
+            num_frames (`int`, defaults to `129`):
+                The number of frames in the generated video.
+            num_inference_steps (`int`, defaults to `50`):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
+            true_cfg_scale (`float`, *optional*, defaults to 1.0):
+                True classifier-free guidance (guidance scale) is enabled when `true_cfg_scale` > 1 and
+                `negative_prompt` is provided.
+            guidance_scale (`float`, defaults to `6.0`):
+                Embedded guiddance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+                a model to generate images more aligned with `prompt` at the expense of lower image quality.
+
+                Guidance-distilled models approximates true classifer-free guidance for `guidance_scale` > 1. Refer to
+                the [paper](https://huggingface.co/papers/2210.03142) to learn more.
+            num_videos_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            latents (`torch.Tensor`, *optional*):
+                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor is generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.Tensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+                input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`HunyuanVideoPipelineOutput`] instead of a plain tuple.
+            attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
+            callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+                A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+                each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+                DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+                list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeline class.
+
+        Examples:
+
+        Returns:
+            [`~HunyuanVideoPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`HunyuanVideoPipelineOutput`] is returned, otherwise a `tuple` is returned
+                where the first element is a list with the generated images and the second element is a list of `bool`s
+                indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw) content.
+        """
+
+        if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+            callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            prompt_2,
+            height,
+            width,
+            prompt_embeds,
+            callback_on_step_end_tensor_inputs,
+            prompt_template,
+        )
+
+        has_neg_prompt = negative_prompt is not None or (
+            negative_prompt_embeds is not None and negative_pooled_prompt_embeds is not None
+        )
+        do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
+
+        self._guidance_scale = guidance_scale
+        self._attention_kwargs = attention_kwargs
+        self._current_timestep = None
+        self._interrupt = False
+
+        device = self._execution_device
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        # 3. Encode input prompt
+        transformer_dtype = self.transformer.dtype
+        prompt_embeds, pooled_prompt_embeds, prompt_attention_mask = self.encode_prompt(
+            prompt=prompt,
+            prompt_2=prompt_2,
+            prompt_template=prompt_template,
+            num_videos_per_prompt=num_videos_per_prompt,
+            prompt_embeds=prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            prompt_attention_mask=prompt_attention_mask,
+            device=device,
+            max_sequence_length=max_sequence_length,
+        )
+        prompt_embeds = prompt_embeds.to(transformer_dtype)
+        prompt_attention_mask = prompt_attention_mask.to(transformer_dtype)
+        pooled_prompt_embeds = pooled_prompt_embeds.to(transformer_dtype)
+
+        if do_true_cfg:
+            negative_prompt_embeds, negative_pooled_prompt_embeds, negative_prompt_attention_mask = self.encode_prompt(
+                prompt=negative_prompt,
+                prompt_2=negative_prompt_2,
+                prompt_template=prompt_template,
+                num_videos_per_prompt=num_videos_per_prompt,
+                prompt_embeds=negative_prompt_embeds,
+                pooled_prompt_embeds=negative_pooled_prompt_embeds,
+                prompt_attention_mask=negative_prompt_attention_mask,
+                device=device,
+                max_sequence_length=max_sequence_length,
+            )
+            negative_prompt_embeds = negative_prompt_embeds.to(transformer_dtype)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.to(transformer_dtype)
+            negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.to(transformer_dtype)
+
+        # 4. Prepare timesteps
+        sigmas = np.linspace(1.0, 0.0, num_inference_steps + 1)[:-1] if sigmas is None else sigmas
+        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, sigmas=sigmas)
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.transformer.config.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_videos_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            num_frames,
+            torch.float32,
+            device,
+            generator,
+            latents,
+        )
+
+        # 6. Prepare guidance condition
+        guidance = torch.tensor([guidance_scale] * latents.shape[0], dtype=transformer_dtype, device=device) * 1000.0
+
+        # 7. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        self._num_timesteps = len(timesteps)
+
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if self.interrupt:
+                    continue
+
+                self._current_timestep = t
+                latent_model_input = latents.to(transformer_dtype)
+                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+                timestep = t.expand(latents.shape[0]).to(latents.dtype)
+
+                with self.transformer.cache_context("cond"):
+                    noise_pred = self.transformer(
+                        hidden_states=latent_model_input,
+                        timestep=timestep,
+                        encoder_hidden_states=prompt_embeds,
+                        encoder_attention_mask=prompt_attention_mask,
+                        pooled_projections=pooled_prompt_embeds,
+                        guidance=guidance,
+                        attention_kwargs=attention_kwargs,
+                        return_dict=False,
+                    )[0]
+
+                if do_true_cfg:
+                    with self.transformer.cache_context("uncond"):
+                        neg_noise_pred = self.transformer(
+                            hidden_states=latent_model_input,
+                            timestep=timestep,
+                            encoder_hidden_states=negative_prompt_embeds,
+                            encoder_attention_mask=negative_prompt_attention_mask,
+                            pooled_projections=negative_pooled_prompt_embeds,
+                            guidance=guidance,
+                            attention_kwargs=attention_kwargs,
+                            return_dict=False,
+                        )[0]
+                    noise_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
+        self._current_timestep = None
+
+        if not output_type == "latent":
+            latents = latents.to(self.vae.dtype) / self.vae.config.scaling_factor
+            video = self.vae.decode(latents, return_dict=False)[0]
+            video = self.video_processor.postprocess_video(video, output_type=output_type)
+        else:
+            video = latents
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (video,)
+
+        return HunyuanVideoPipelineOutput(frames=video)

pythonProject/.venv/Lib/site-packages/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video_framepack.py

@@ -0,0 +1,1114 @@
+# Copyright 2025 The Framepack Team, The HunyuanVideo Team and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+import math
+from enum import Enum
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+from transformers import (
+    CLIPTextModel,
+    CLIPTokenizer,
+    LlamaModel,
+    LlamaTokenizerFast,
+    SiglipImageProcessor,
+    SiglipVisionModel,
+)
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput
+from ...loaders import HunyuanVideoLoraLoaderMixin
+from ...models import AutoencoderKLHunyuanVideo, HunyuanVideoFramepackTransformer3DModel
+from ...schedulers import FlowMatchEulerDiscreteScheduler
+from ...utils import is_torch_xla_available, logging, replace_example_docstring
+from ...utils.torch_utils import randn_tensor
+from ...video_processor import VideoProcessor
+from ..pipeline_utils import DiffusionPipeline
+from .pipeline_output import HunyuanVideoFramepackPipelineOutput
+
+
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+# TODO(yiyi): We can pack the checkpoints nicely with modular loader
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ##### Image-to-Video
+
+        ```python
+        >>> import torch
+        >>> from diffusers import HunyuanVideoFramepackPipeline, HunyuanVideoFramepackTransformer3DModel
+        >>> from diffusers.utils import export_to_video, load_image
+        >>> from transformers import SiglipImageProcessor, SiglipVisionModel
+
+        >>> transformer = HunyuanVideoFramepackTransformer3DModel.from_pretrained(
+        ...     "lllyasviel/FramePackI2V_HY", torch_dtype=torch.bfloat16
+        ... )
+        >>> feature_extractor = SiglipImageProcessor.from_pretrained(
+        ...     "lllyasviel/flux_redux_bfl", subfolder="feature_extractor"
+        ... )
+        >>> image_encoder = SiglipVisionModel.from_pretrained(
+        ...     "lllyasviel/flux_redux_bfl", subfolder="image_encoder", torch_dtype=torch.float16
+        ... )
+        >>> pipe = HunyuanVideoFramepackPipeline.from_pretrained(
+        ...     "hunyuanvideo-community/HunyuanVideo",
+        ...     transformer=transformer,
+        ...     feature_extractor=feature_extractor,
+        ...     image_encoder=image_encoder,
+        ...     torch_dtype=torch.float16,
+        ... )
+        >>> pipe.vae.enable_tiling()
+        >>> pipe.to("cuda")
+
+        >>> image = load_image(
+        ...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/penguin.png"
+        ... )
+        >>> output = pipe(
+        ...     image=image,
+        ...     prompt="A penguin dancing in the snow",
+        ...     height=832,
+        ...     width=480,
+        ...     num_frames=91,
+        ...     num_inference_steps=30,
+        ...     guidance_scale=9.0,
+        ...     generator=torch.Generator().manual_seed(0),
+        ...     sampling_type="inverted_anti_drifting",
+        ... ).frames[0]
+        >>> export_to_video(output, "output.mp4", fps=30)
+        ```
+
+        ##### First and Last Image-to-Video
+
+        ```python
+        >>> import torch
+        >>> from diffusers import HunyuanVideoFramepackPipeline, HunyuanVideoFramepackTransformer3DModel
+        >>> from diffusers.utils import export_to_video, load_image
+        >>> from transformers import SiglipImageProcessor, SiglipVisionModel
+
+        >>> transformer = HunyuanVideoFramepackTransformer3DModel.from_pretrained(
+        ...     "lllyasviel/FramePackI2V_HY", torch_dtype=torch.bfloat16
+        ... )
+        >>> feature_extractor = SiglipImageProcessor.from_pretrained(
+        ...     "lllyasviel/flux_redux_bfl", subfolder="feature_extractor"
+        ... )
+        >>> image_encoder = SiglipVisionModel.from_pretrained(
+        ...     "lllyasviel/flux_redux_bfl", subfolder="image_encoder", torch_dtype=torch.float16
+        ... )
+        >>> pipe = HunyuanVideoFramepackPipeline.from_pretrained(
+        ...     "hunyuanvideo-community/HunyuanVideo",
+        ...     transformer=transformer,
+        ...     feature_extractor=feature_extractor,
+        ...     image_encoder=image_encoder,
+        ...     torch_dtype=torch.float16,
+        ... )
+        >>> pipe.to("cuda")
+
+        >>> prompt = "CG animation style, a small blue bird takes off from the ground, flapping its wings. The bird's feathers are delicate, with a unique pattern on its chest. The background shows a blue sky with white clouds under bright sunshine. The camera follows the bird upward, capturing its flight and the vastness of the sky from a close-up, low-angle perspective."
+        >>> first_image = load_image(
+        ...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/flf2v_input_first_frame.png"
+        ... )
+        >>> last_image = load_image(
+        ...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/flf2v_input_last_frame.png"
+        ... )
+        >>> output = pipe(
+        ...     image=first_image,
+        ...     last_image=last_image,
+        ...     prompt=prompt,
+        ...     height=512,
+        ...     width=512,
+        ...     num_frames=91,
+        ...     num_inference_steps=30,
+        ...     guidance_scale=9.0,
+        ...     generator=torch.Generator().manual_seed(0),
+        ...     sampling_type="inverted_anti_drifting",
+        ... ).frames[0]
+        >>> export_to_video(output, "output.mp4", fps=30)
+        ```
+"""
+
+
+DEFAULT_PROMPT_TEMPLATE = {
+    "template": (
+        "<|start_header_id|>system<|end_header_id|>\n\nDescribe the video by detailing the following aspects: "
+        "1. The main content and theme of the video."
+        "2. The color, shape, size, texture, quantity, text, and spatial relationships of the objects."
+        "3. Actions, events, behaviors temporal relationships, physical movement changes of the objects."
+        "4. background environment, light, style and atmosphere."
+        "5. camera angles, movements, and transitions used in the video:<|eot_id|>"
+        "<|start_header_id|>user<|end_header_id|>\n\n{}<|eot_id|>"
+    ),
+    "crop_start": 95,
+}
+
+
+# Copied from diffusers.pipelines.flux.pipeline_flux.calculate_shift
+def calculate_shift(
+    image_seq_len,
+    base_seq_len: int = 256,
+    max_seq_len: int = 4096,
+    base_shift: float = 0.5,
+    max_shift: float = 1.15,
+):
+    m = (max_shift - base_shift) / (max_seq_len - base_seq_len)
+    b = base_shift - m * base_seq_len
+    mu = image_seq_len * m + b
+    return mu
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
+    **kwargs,
+):
+    r"""
+    Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+    custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+            must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
+class FramepackSamplingType(str, Enum):
+    VANILLA = "vanilla"
+    INVERTED_ANTI_DRIFTING = "inverted_anti_drifting"
+
+
+class HunyuanVideoFramepackPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
+    r"""
+    Pipeline for text-to-video generation using HunyuanVideo.
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+    implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+    Args:
+        text_encoder ([`LlamaModel`]):
+            [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
+        tokenizer (`LlamaTokenizer`):
+            Tokenizer from [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
+        transformer ([`HunyuanVideoTransformer3DModel`]):
+            Conditional Transformer to denoise the encoded image latents.
+        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
+            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
+        vae ([`AutoencoderKLHunyuanVideo`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
+        text_encoder_2 ([`CLIPTextModel`]):
+            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+        tokenizer_2 (`CLIPTokenizer`):
+            Tokenizer of class
+            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
+    """
+
+    model_cpu_offload_seq = "text_encoder->text_encoder_2->image_encoder->transformer->vae"
+    _callback_tensor_inputs = ["latents", "prompt_embeds"]
+
+    def __init__(
+        self,
+        text_encoder: LlamaModel,
+        tokenizer: LlamaTokenizerFast,
+        transformer: HunyuanVideoFramepackTransformer3DModel,
+        vae: AutoencoderKLHunyuanVideo,
+        scheduler: FlowMatchEulerDiscreteScheduler,
+        text_encoder_2: CLIPTextModel,
+        tokenizer_2: CLIPTokenizer,
+        image_encoder: SiglipVisionModel,
+        feature_extractor: SiglipImageProcessor,
+    ):
+        super().__init__()
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            transformer=transformer,
+            scheduler=scheduler,
+            text_encoder_2=text_encoder_2,
+            tokenizer_2=tokenizer_2,
+            image_encoder=image_encoder,
+            feature_extractor=feature_extractor,
+        )
+
+        self.vae_scale_factor_temporal = self.vae.temporal_compression_ratio if getattr(self, "vae", None) else 4
+        self.vae_scale_factor_spatial = self.vae.spatial_compression_ratio if getattr(self, "vae", None) else 8
+        self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline._get_llama_prompt_embeds
+    def _get_llama_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]],
+        prompt_template: Dict[str, Any],
+        num_videos_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 256,
+        num_hidden_layers_to_skip: int = 2,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt)
+
+        prompt = [prompt_template["template"].format(p) for p in prompt]
+
+        crop_start = prompt_template.get("crop_start", None)
+        if crop_start is None:
+            prompt_template_input = self.tokenizer(
+                prompt_template["template"],
+                padding="max_length",
+                return_tensors="pt",
+                return_length=False,
+                return_overflowing_tokens=False,
+                return_attention_mask=False,
+            )
+            crop_start = prompt_template_input["input_ids"].shape[-1]
+            # Remove <|eot_id|> token and placeholder {}
+            crop_start -= 2
+
+        max_sequence_length += crop_start
+        text_inputs = self.tokenizer(
+            prompt,
+            max_length=max_sequence_length,
+            padding="max_length",
+            truncation=True,
+            return_tensors="pt",
+            return_length=False,
+            return_overflowing_tokens=False,
+            return_attention_mask=True,
+        )
+        text_input_ids = text_inputs.input_ids.to(device=device)
+        prompt_attention_mask = text_inputs.attention_mask.to(device=device)
+
+        prompt_embeds = self.text_encoder(
+            input_ids=text_input_ids,
+            attention_mask=prompt_attention_mask,
+            output_hidden_states=True,
+        ).hidden_states[-(num_hidden_layers_to_skip + 1)]
+        prompt_embeds = prompt_embeds.to(dtype=dtype)
+
+        if crop_start is not None and crop_start > 0:
+            prompt_embeds = prompt_embeds[:, crop_start:]
+            prompt_attention_mask = prompt_attention_mask[:, crop_start:]
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        _, seq_len, _ = prompt_embeds.shape
+        prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1)
+        prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)
+        prompt_attention_mask = prompt_attention_mask.repeat(1, num_videos_per_prompt)
+        prompt_attention_mask = prompt_attention_mask.view(batch_size * num_videos_per_prompt, seq_len)
+
+        return prompt_embeds, prompt_attention_mask
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline._get_clip_prompt_embeds
+    def _get_clip_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]],
+        num_videos_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 77,
+    ) -> torch.Tensor:
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder_2.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt)
+
+        text_inputs = self.tokenizer_2(
+            prompt,
+            padding="max_length",
+            max_length=max_sequence_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+
+        text_input_ids = text_inputs.input_ids
+        untruncated_ids = self.tokenizer_2(prompt, padding="longest", return_tensors="pt").input_ids
+        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+            removed_text = self.tokenizer_2.batch_decode(untruncated_ids[:, max_sequence_length - 1 : -1])
+            logger.warning(
+                "The following part of your input was truncated because CLIP can only handle sequences up to"
+                f" {max_sequence_length} tokens: {removed_text}"
+            )
+
+        prompt_embeds = self.text_encoder_2(text_input_ids.to(device), output_hidden_states=False).pooler_output
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt)
+        prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, -1)
+
+        return prompt_embeds
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline.encode_prompt
+    def encode_prompt(
+        self,
+        prompt: Union[str, List[str]],
+        prompt_2: Union[str, List[str]] = None,
+        prompt_template: Dict[str, Any] = DEFAULT_PROMPT_TEMPLATE,
+        num_videos_per_prompt: int = 1,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_attention_mask: Optional[torch.Tensor] = None,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 256,
+    ):
+        if prompt_embeds is None:
+            prompt_embeds, prompt_attention_mask = self._get_llama_prompt_embeds(
+                prompt,
+                prompt_template,
+                num_videos_per_prompt,
+                device=device,
+                dtype=dtype,
+                max_sequence_length=max_sequence_length,
+            )
+
+        if pooled_prompt_embeds is None:
+            if prompt_2 is None:
+                prompt_2 = prompt
+            pooled_prompt_embeds = self._get_clip_prompt_embeds(
+                prompt,
+                num_videos_per_prompt,
+                device=device,
+                dtype=dtype,
+                max_sequence_length=77,
+            )
+
+        return prompt_embeds, pooled_prompt_embeds, prompt_attention_mask
+
+    def encode_image(
+        self, image: torch.Tensor, device: Optional[torch.device] = None, dtype: Optional[torch.dtype] = None
+    ):
+        device = device or self._execution_device
+        image = (image + 1) / 2.0  # [-1, 1] -> [0, 1]
+        image = self.feature_extractor(images=image, return_tensors="pt", do_rescale=False).to(
+            device=device, dtype=self.image_encoder.dtype
+        )
+        image_embeds = self.image_encoder(**image).last_hidden_state
+        return image_embeds.to(dtype=dtype)
+
+    def check_inputs(
+        self,
+        prompt,
+        prompt_2,
+        height,
+        width,
+        prompt_embeds=None,
+        callback_on_step_end_tensor_inputs=None,
+        prompt_template=None,
+        image=None,
+        image_latents=None,
+        last_image=None,
+        last_image_latents=None,
+        sampling_type=None,
+    ):
+        if height % 16 != 0 or width % 16 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 16 but are {height} and {width}.")
+
+        if callback_on_step_end_tensor_inputs is not None and not all(
+            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+        ):
+            raise ValueError(
+                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+            )
+
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt_2 is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+        elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+            raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+        if prompt_template is not None:
+            if not isinstance(prompt_template, dict):
+                raise ValueError(f"`prompt_template` has to be of type `dict` but is {type(prompt_template)}")
+            if "template" not in prompt_template:
+                raise ValueError(
+                    f"`prompt_template` has to contain a key `template` but only found {prompt_template.keys()}"
+                )
+
+        sampling_types = [x.value for x in FramepackSamplingType.__members__.values()]
+        if sampling_type not in sampling_types:
+            raise ValueError(f"`sampling_type` has to be one of '{sampling_types}' but is '{sampling_type}'")
+
+        if image is not None and image_latents is not None:
+            raise ValueError("Only one of `image` or `image_latents` can be passed.")
+        if last_image is not None and last_image_latents is not None:
+            raise ValueError("Only one of `last_image` or `last_image_latents` can be passed.")
+        if sampling_type != FramepackSamplingType.INVERTED_ANTI_DRIFTING and (
+            last_image is not None or last_image_latents is not None
+        ):
+            raise ValueError(
+                'Only `"inverted_anti_drifting"` inference type supports `last_image` or `last_image_latents`.'
+            )
+
+    def prepare_latents(
+        self,
+        batch_size: int = 1,
+        num_channels_latents: int = 16,
+        height: int = 720,
+        width: int = 1280,
+        num_frames: int = 129,
+        dtype: Optional[torch.dtype] = None,
+        device: Optional[torch.device] = None,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        if latents is not None:
+            return latents.to(device=device, dtype=dtype)
+        shape = (
+            batch_size,
+            num_channels_latents,
+            (num_frames - 1) // self.vae_scale_factor_temporal + 1,
+            int(height) // self.vae_scale_factor_spatial,
+            int(width) // self.vae_scale_factor_spatial,
+        )
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+        latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+        return latents
+
+    def prepare_image_latents(
+        self,
+        image: torch.Tensor,
+        dtype: Optional[torch.dtype] = None,
+        device: Optional[torch.device] = None,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        device = device or self._execution_device
+        if latents is None:
+            image = image.unsqueeze(2).to(device=device, dtype=self.vae.dtype)
+            latents = self.vae.encode(image).latent_dist.sample(generator=generator)
+            latents = latents * self.vae.config.scaling_factor
+        return latents.to(device=device, dtype=dtype)
+
+    def enable_vae_slicing(self):
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.vae.enable_slicing()
+
+    def disable_vae_slicing(self):
+        r"""
+        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_slicing()
+
+    def enable_vae_tiling(self):
+        r"""
+        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+        processing larger images.
+        """
+        self.vae.enable_tiling()
+
+    def disable_vae_tiling(self):
+        r"""
+        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_tiling()
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    @property
+    def num_timesteps(self):
+        return self._num_timesteps
+
+    @property
+    def attention_kwargs(self):
+        return self._attention_kwargs
+
+    @property
+    def current_timestep(self):
+        return self._current_timestep
+
+    @property
+    def interrupt(self):
+        return self._interrupt
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        image: PipelineImageInput,
+        last_image: Optional[PipelineImageInput] = None,
+        prompt: Union[str, List[str]] = None,
+        prompt_2: Union[str, List[str]] = None,
+        negative_prompt: Union[str, List[str]] = None,
+        negative_prompt_2: Union[str, List[str]] = None,
+        height: int = 720,
+        width: int = 1280,
+        num_frames: int = 129,
+        latent_window_size: int = 9,
+        num_inference_steps: int = 50,
+        sigmas: List[float] = None,
+        true_cfg_scale: float = 1.0,
+        guidance_scale: float = 6.0,
+        num_videos_per_prompt: Optional[int] = 1,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        image_latents: Optional[torch.Tensor] = None,
+        last_image_latents: Optional[torch.Tensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_attention_mask: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        attention_kwargs: Optional[Dict[str, Any]] = None,
+        callback_on_step_end: Optional[
+            Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+        ] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        prompt_template: Dict[str, Any] = DEFAULT_PROMPT_TEMPLATE,
+        max_sequence_length: int = 256,
+        sampling_type: FramepackSamplingType = FramepackSamplingType.INVERTED_ANTI_DRIFTING,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor`):
+                The image to be used as the starting point for the video generation.
+            last_image (`PIL.Image.Image` or `np.ndarray` or `torch.Tensor`, *optional*):
+                The optional last image to be used as the ending point for the video generation. This is useful for
+                generating transitions between two images.
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+                will be used instead.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
+                not greater than `1`).
+            negative_prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+                `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
+            height (`int`, defaults to `720`):
+                The height in pixels of the generated image.
+            width (`int`, defaults to `1280`):
+                The width in pixels of the generated image.
+            num_frames (`int`, defaults to `129`):
+                The number of frames in the generated video.
+            num_inference_steps (`int`, defaults to `50`):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
+            true_cfg_scale (`float`, *optional*, defaults to 1.0):
+                When > 1.0 and a provided `negative_prompt`, enables true classifier-free guidance.
+            guidance_scale (`float`, defaults to `6.0`):
+                Guidance scale as defined in [Classifier-Free Diffusion
+                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+                the text `prompt`, usually at the expense of lower image quality. Note that the only available
+                HunyuanVideo model is CFG-distilled, which means that traditional guidance between unconditional and
+                conditional latent is not applied.
+            num_videos_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            image_latents (`torch.Tensor`, *optional*):
+                Pre-encoded image latents. If not provided, the image will be encoded using the VAE.
+            last_image_latents (`torch.Tensor`, *optional*):
+                Pre-encoded last image latents. If not provided, the last image will be encoded using the VAE.
+            prompt_embeds (`torch.Tensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+                input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`HunyuanVideoFramepackPipelineOutput`] instead of a plain tuple.
+            attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
+            callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+                A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+                each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+                DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+                list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeline class.
+
+        Examples:
+
+        Returns:
+            [`~HunyuanVideoFramepackPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`HunyuanVideoFramepackPipelineOutput`] is returned, otherwise a `tuple` is
+                returned where the first element is a list with the generated images and the second element is a list
+                of `bool`s indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw)
+                content.
+        """
+
+        if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+            callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            prompt_2,
+            height,
+            width,
+            prompt_embeds,
+            callback_on_step_end_tensor_inputs,
+            prompt_template,
+            image,
+            image_latents,
+            last_image,
+            last_image_latents,
+            sampling_type,
+        )
+
+        has_neg_prompt = negative_prompt is not None or (
+            negative_prompt_embeds is not None and negative_pooled_prompt_embeds is not None
+        )
+        do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
+
+        self._guidance_scale = guidance_scale
+        self._attention_kwargs = attention_kwargs
+        self._current_timestep = None
+        self._interrupt = False
+
+        device = self._execution_device
+        transformer_dtype = self.transformer.dtype
+        vae_dtype = self.vae.dtype
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        # 3. Encode input prompt
+        transformer_dtype = self.transformer.dtype
+        prompt_embeds, pooled_prompt_embeds, prompt_attention_mask = self.encode_prompt(
+            prompt=prompt,
+            prompt_2=prompt_2,
+            prompt_template=prompt_template,
+            num_videos_per_prompt=num_videos_per_prompt,
+            prompt_embeds=prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            prompt_attention_mask=prompt_attention_mask,
+            device=device,
+            max_sequence_length=max_sequence_length,
+        )
+        prompt_embeds = prompt_embeds.to(transformer_dtype)
+        prompt_attention_mask = prompt_attention_mask.to(transformer_dtype)
+        pooled_prompt_embeds = pooled_prompt_embeds.to(transformer_dtype)
+
+        if do_true_cfg:
+            negative_prompt_embeds, negative_pooled_prompt_embeds, negative_prompt_attention_mask = self.encode_prompt(
+                prompt=negative_prompt,
+                prompt_2=negative_prompt_2,
+                prompt_template=prompt_template,
+                num_videos_per_prompt=num_videos_per_prompt,
+                prompt_embeds=negative_prompt_embeds,
+                pooled_prompt_embeds=negative_pooled_prompt_embeds,
+                prompt_attention_mask=negative_prompt_attention_mask,
+                device=device,
+                max_sequence_length=max_sequence_length,
+            )
+            negative_prompt_embeds = negative_prompt_embeds.to(transformer_dtype)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.to(transformer_dtype)
+            negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.to(transformer_dtype)
+
+        # 4. Prepare image
+        image = self.video_processor.preprocess(image, height, width)
+        image_embeds = self.encode_image(image, device=device).to(transformer_dtype)
+        if last_image is not None:
+            # Credits: https://github.com/lllyasviel/FramePack/pull/167
+            # Users can modify the weighting strategy applied here
+            last_image = self.video_processor.preprocess(last_image, height, width)
+            last_image_embeds = self.encode_image(last_image, device=device).to(transformer_dtype)
+            last_image_embeds = (image_embeds + last_image_embeds) / 2
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.transformer.config.in_channels
+        window_num_frames = (latent_window_size - 1) * self.vae_scale_factor_temporal + 1
+        num_latent_sections = max(1, (num_frames + window_num_frames - 1) // window_num_frames)
+        history_video = None
+        total_generated_latent_frames = 0
+
+        image_latents = self.prepare_image_latents(
+            image, dtype=torch.float32, device=device, generator=generator, latents=image_latents
+        )
+        if last_image is not None:
+            last_image_latents = self.prepare_image_latents(
+                last_image, dtype=torch.float32, device=device, generator=generator
+            )
+
+        # Specific to the released checkpoints:
+        #   - https://huggingface.co/lllyasviel/FramePackI2V_HY
+        #   - https://huggingface.co/lllyasviel/FramePack_F1_I2V_HY_20250503
+        # TODO: find a more generic way in future if there are more checkpoints
+        if sampling_type == FramepackSamplingType.INVERTED_ANTI_DRIFTING:
+            history_sizes = [1, 2, 16]
+            history_latents = torch.zeros(
+                batch_size,
+                num_channels_latents,
+                sum(history_sizes),
+                height // self.vae_scale_factor_spatial,
+                width // self.vae_scale_factor_spatial,
+                device=device,
+                dtype=torch.float32,
+            )
+
+        elif sampling_type == FramepackSamplingType.VANILLA:
+            history_sizes = [16, 2, 1]
+            history_latents = torch.zeros(
+                batch_size,
+                num_channels_latents,
+                sum(history_sizes),
+                height // self.vae_scale_factor_spatial,
+                width // self.vae_scale_factor_spatial,
+                device=device,
+                dtype=torch.float32,
+            )
+            history_latents = torch.cat([history_latents, image_latents], dim=2)
+            total_generated_latent_frames += 1
+
+        else:
+            assert False
+
+        # 6. Prepare guidance condition
+        guidance = torch.tensor([guidance_scale] * batch_size, dtype=transformer_dtype, device=device) * 1000.0
+
+        # 7. Denoising loop
+        for k in range(num_latent_sections):
+            if sampling_type == FramepackSamplingType.INVERTED_ANTI_DRIFTING:
+                latent_paddings = list(reversed(range(num_latent_sections)))
+                if num_latent_sections > 4:
+                    latent_paddings = [3] + [2] * (num_latent_sections - 3) + [1, 0]
+
+                is_first_section = k == 0
+                is_last_section = k == num_latent_sections - 1
+                latent_padding_size = latent_paddings[k] * latent_window_size
+
+                indices = torch.arange(0, sum([1, latent_padding_size, latent_window_size, *history_sizes]))
+                (
+                    indices_prefix,
+                    indices_padding,
+                    indices_latents,
+                    indices_latents_history_1x,
+                    indices_latents_history_2x,
+                    indices_latents_history_4x,
+                ) = indices.split([1, latent_padding_size, latent_window_size, *history_sizes], dim=0)
+                # Inverted anti-drifting sampling: Figure 2(c) in the paper
+                indices_clean_latents = torch.cat([indices_prefix, indices_latents_history_1x], dim=0)
+
+                latents_prefix = image_latents
+                latents_history_1x, latents_history_2x, latents_history_4x = history_latents[
+                    :, :, : sum(history_sizes)
+                ].split(history_sizes, dim=2)
+                if last_image is not None and is_first_section:
+                    latents_history_1x = last_image_latents
+                latents_clean = torch.cat([latents_prefix, latents_history_1x], dim=2)
+
+            elif sampling_type == FramepackSamplingType.VANILLA:
+                indices = torch.arange(0, sum([1, *history_sizes, latent_window_size]))
+                (
+                    indices_prefix,
+                    indices_latents_history_4x,
+                    indices_latents_history_2x,
+                    indices_latents_history_1x,
+                    indices_latents,
+                ) = indices.split([1, *history_sizes, latent_window_size], dim=0)
+                indices_clean_latents = torch.cat([indices_prefix, indices_latents_history_1x], dim=0)
+
+                latents_prefix = image_latents
+                latents_history_4x, latents_history_2x, latents_history_1x = history_latents[
+                    :, :, -sum(history_sizes) :
+                ].split(history_sizes, dim=2)
+                latents_clean = torch.cat([latents_prefix, latents_history_1x], dim=2)
+
+            else:
+                assert False
+
+            latents = self.prepare_latents(
+                batch_size,
+                num_channels_latents,
+                height,
+                width,
+                window_num_frames,
+                dtype=torch.float32,
+                device=device,
+                generator=generator,
+                latents=None,
+            )
+
+            sigmas = np.linspace(1.0, 0.0, num_inference_steps + 1)[:-1] if sigmas is None else sigmas
+            image_seq_len = (
+                latents.shape[2] * latents.shape[3] * latents.shape[4] / self.transformer.config.patch_size**2
+            )
+            exp_max = 7.0
+            mu = calculate_shift(
+                image_seq_len,
+                self.scheduler.config.get("base_image_seq_len", 256),
+                self.scheduler.config.get("max_image_seq_len", 4096),
+                self.scheduler.config.get("base_shift", 0.5),
+                self.scheduler.config.get("max_shift", 1.15),
+            )
+            mu = min(mu, math.log(exp_max))
+            timesteps, num_inference_steps = retrieve_timesteps(
+                self.scheduler, num_inference_steps, device, sigmas=sigmas, mu=mu
+            )
+            num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+            self._num_timesteps = len(timesteps)
+
+            with self.progress_bar(total=num_inference_steps) as progress_bar:
+                for i, t in enumerate(timesteps):
+                    if self.interrupt:
+                        continue
+
+                    self._current_timestep = t
+                    timestep = t.expand(latents.shape[0])
+
+                    noise_pred = self.transformer(
+                        hidden_states=latents.to(transformer_dtype),
+                        timestep=timestep,
+                        encoder_hidden_states=prompt_embeds,
+                        encoder_attention_mask=prompt_attention_mask,
+                        pooled_projections=pooled_prompt_embeds,
+                        image_embeds=image_embeds,
+                        indices_latents=indices_latents,
+                        guidance=guidance,
+                        latents_clean=latents_clean.to(transformer_dtype),
+                        indices_latents_clean=indices_clean_latents,
+                        latents_history_2x=latents_history_2x.to(transformer_dtype),
+                        indices_latents_history_2x=indices_latents_history_2x,
+                        latents_history_4x=latents_history_4x.to(transformer_dtype),
+                        indices_latents_history_4x=indices_latents_history_4x,
+                        attention_kwargs=attention_kwargs,
+                        return_dict=False,
+                    )[0]
+
+                    if do_true_cfg:
+                        neg_noise_pred = self.transformer(
+                            hidden_states=latents.to(transformer_dtype),
+                            timestep=timestep,
+                            encoder_hidden_states=negative_prompt_embeds,
+                            encoder_attention_mask=negative_prompt_attention_mask,
+                            pooled_projections=negative_pooled_prompt_embeds,
+                            image_embeds=image_embeds,
+                            indices_latents=indices_latents,
+                            guidance=guidance,
+                            latents_clean=latents_clean.to(transformer_dtype),
+                            indices_latents_clean=indices_clean_latents,
+                            latents_history_2x=latents_history_2x.to(transformer_dtype),
+                            indices_latents_history_2x=indices_latents_history_2x,
+                            latents_history_4x=latents_history_4x.to(transformer_dtype),
+                            indices_latents_history_4x=indices_latents_history_4x,
+                            attention_kwargs=attention_kwargs,
+                            return_dict=False,
+                        )[0]
+                        noise_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)
+
+                    # compute the previous noisy sample x_t -> x_t-1
+                    latents = self.scheduler.step(noise_pred.float(), t, latents, return_dict=False)[0]
+
+                    if callback_on_step_end is not None:
+                        callback_kwargs = {}
+                        for k in callback_on_step_end_tensor_inputs:
+                            callback_kwargs[k] = locals()[k]
+                        callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                        latents = callback_outputs.pop("latents", latents)
+                        prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
+                    # call the callback, if provided
+                    if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                        progress_bar.update()
+
+                    if XLA_AVAILABLE:
+                        xm.mark_step()
+
+                if sampling_type == FramepackSamplingType.INVERTED_ANTI_DRIFTING:
+                    if is_last_section:
+                        latents = torch.cat([image_latents, latents], dim=2)
+                    total_generated_latent_frames += latents.shape[2]
+                    history_latents = torch.cat([latents, history_latents], dim=2)
+                    real_history_latents = history_latents[:, :, :total_generated_latent_frames]
+                    section_latent_frames = (
+                        (latent_window_size * 2 + 1) if is_last_section else (latent_window_size * 2)
+                    )
+                    index_slice = (slice(None), slice(None), slice(0, section_latent_frames))
+
+                elif sampling_type == FramepackSamplingType.VANILLA:
+                    total_generated_latent_frames += latents.shape[2]
+                    history_latents = torch.cat([history_latents, latents], dim=2)
+                    real_history_latents = history_latents[:, :, -total_generated_latent_frames:]
+                    section_latent_frames = latent_window_size * 2
+                    index_slice = (slice(None), slice(None), slice(-section_latent_frames, None))
+
+                else:
+                    assert False
+
+                if history_video is None:
+                    if not output_type == "latent":
+                        current_latents = real_history_latents.to(vae_dtype) / self.vae.config.scaling_factor
+                        history_video = self.vae.decode(current_latents, return_dict=False)[0]
+                    else:
+                        history_video = [real_history_latents]
+                else:
+                    if not output_type == "latent":
+                        overlapped_frames = (latent_window_size - 1) * self.vae_scale_factor_temporal + 1
+                        current_latents = (
+                            real_history_latents[index_slice].to(vae_dtype) / self.vae.config.scaling_factor
+                        )
+                        current_video = self.vae.decode(current_latents, return_dict=False)[0]
+
+                        if sampling_type == FramepackSamplingType.INVERTED_ANTI_DRIFTING:
+                            history_video = self._soft_append(current_video, history_video, overlapped_frames)
+                        elif sampling_type == FramepackSamplingType.VANILLA:
+                            history_video = self._soft_append(history_video, current_video, overlapped_frames)
+                        else:
+                            assert False
+                    else:
+                        history_video.append(real_history_latents)
+
+        self._current_timestep = None
+
+        if not output_type == "latent":
+            generated_frames = history_video.size(2)
+            generated_frames = (
+                generated_frames - 1
+            ) // self.vae_scale_factor_temporal * self.vae_scale_factor_temporal + 1
+            history_video = history_video[:, :, :generated_frames]
+            video = self.video_processor.postprocess_video(history_video, output_type=output_type)
+        else:
+            video = history_video
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (video,)
+
+        return HunyuanVideoFramepackPipelineOutput(frames=video)
+
+    def _soft_append(self, history: torch.Tensor, current: torch.Tensor, overlap: int = 0):
+        if overlap <= 0:
+            return torch.cat([history, current], dim=2)
+
+        assert history.shape[2] >= overlap, f"Current length ({history.shape[2]}) must be >= overlap ({overlap})"
+        assert current.shape[2] >= overlap, f"History length ({current.shape[2]}) must be >= overlap ({overlap})"
+
+        weights = torch.linspace(1, 0, overlap, dtype=history.dtype, device=history.device).view(1, 1, -1, 1, 1)
+        blended = weights * history[:, :, -overlap:] + (1 - weights) * current[:, :, :overlap]
+        output = torch.cat([history[:, :, :-overlap], blended, current[:, :, overlap:]], dim=2)
+
+        return output.to(history)

pythonProject/.venv/Lib/site-packages/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video_image2video.py

@@ -0,0 +1,980 @@
+# Copyright 2025 The HunyuanVideo Team and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+from transformers import (
+    CLIPImageProcessor,
+    CLIPTextModel,
+    CLIPTokenizer,
+    LlamaTokenizerFast,
+    LlavaForConditionalGeneration,
+)
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...loaders import HunyuanVideoLoraLoaderMixin
+from ...models import AutoencoderKLHunyuanVideo, HunyuanVideoTransformer3DModel
+from ...schedulers import FlowMatchEulerDiscreteScheduler
+from ...utils import is_torch_xla_available, logging, replace_example_docstring
+from ...utils.torch_utils import randn_tensor
+from ...video_processor import VideoProcessor
+from ..pipeline_utils import DiffusionPipeline
+from .pipeline_output import HunyuanVideoPipelineOutput
+
+
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```python
+        >>> import torch
+        >>> from diffusers import HunyuanVideoImageToVideoPipeline, HunyuanVideoTransformer3DModel
+        >>> from diffusers.utils import load_image, export_to_video
+
+        >>> # Available checkpoints: hunyuanvideo-community/HunyuanVideo-I2V, hunyuanvideo-community/HunyuanVideo-I2V-33ch
+        >>> model_id = "hunyuanvideo-community/HunyuanVideo-I2V"
+        >>> transformer = HunyuanVideoTransformer3DModel.from_pretrained(
+        ...     model_id, subfolder="transformer", torch_dtype=torch.bfloat16
+        ... )
+        >>> pipe = HunyuanVideoImageToVideoPipeline.from_pretrained(
+        ...     model_id, transformer=transformer, torch_dtype=torch.float16
+        ... )
+        >>> pipe.vae.enable_tiling()
+        >>> pipe.to("cuda")
+
+        >>> prompt = "A man with short gray hair plays a red electric guitar."
+        >>> image = load_image(
+        ...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/guitar-man.png"
+        ... )
+
+        >>> # If using hunyuanvideo-community/HunyuanVideo-I2V
+        >>> output = pipe(image=image, prompt=prompt, guidance_scale=6.0).frames[0]
+
+        >>> # If using hunyuanvideo-community/HunyuanVideo-I2V-33ch
+        >>> output = pipe(image=image, prompt=prompt, guidance_scale=1.0, true_cfg_scale=1.0).frames[0]
+
+        >>> export_to_video(output, "output.mp4", fps=15)
+        ```
+"""
+
+
+DEFAULT_PROMPT_TEMPLATE = {
+    "template": (
+        "<|start_header_id|>system<|end_header_id|>\n\n<image>\nDescribe the video by detailing the following aspects according to the reference image: "
+        "1. The main content and theme of the video."
+        "2. The color, shape, size, texture, quantity, text, and spatial relationships of the objects."
+        "3. Actions, events, behaviors temporal relationships, physical movement changes of the objects."
+        "4. background environment, light, style and atmosphere."
+        "5. camera angles, movements, and transitions used in the video:<|eot_id|>\n\n"
+        "<|start_header_id|>user<|end_header_id|>\n\n{}<|eot_id|>"
+        "<|start_header_id|>assistant<|end_header_id|>\n\n"
+    ),
+    "crop_start": 103,
+    "image_emb_start": 5,
+    "image_emb_end": 581,
+    "image_emb_len": 576,
+    "double_return_token_id": 271,
+}
+
+
+def _expand_input_ids_with_image_tokens(
+    text_input_ids,
+    prompt_attention_mask,
+    max_sequence_length,
+    image_token_index,
+    image_emb_len,
+    image_emb_start,
+    image_emb_end,
+    pad_token_id,
+):
+    special_image_token_mask = text_input_ids == image_token_index
+    num_special_image_tokens = torch.sum(special_image_token_mask, dim=-1)
+    batch_indices, non_image_indices = torch.where(text_input_ids != image_token_index)
+
+    max_expanded_length = max_sequence_length + (num_special_image_tokens.max() * (image_emb_len - 1))
+    new_token_positions = torch.cumsum((special_image_token_mask * (image_emb_len - 1) + 1), -1) - 1
+    text_to_overwrite = new_token_positions[batch_indices, non_image_indices]
+
+    expanded_input_ids = torch.full(
+        (text_input_ids.shape[0], max_expanded_length),
+        pad_token_id,
+        dtype=text_input_ids.dtype,
+        device=text_input_ids.device,
+    )
+    expanded_input_ids[batch_indices, text_to_overwrite] = text_input_ids[batch_indices, non_image_indices]
+    expanded_input_ids[batch_indices, image_emb_start:image_emb_end] = image_token_index
+
+    expanded_attention_mask = torch.zeros(
+        (text_input_ids.shape[0], max_expanded_length),
+        dtype=prompt_attention_mask.dtype,
+        device=prompt_attention_mask.device,
+    )
+    attn_batch_indices, attention_indices = torch.where(expanded_input_ids != pad_token_id)
+    expanded_attention_mask[attn_batch_indices, attention_indices] = 1.0
+    expanded_attention_mask = expanded_attention_mask.to(prompt_attention_mask.dtype)
+    position_ids = (expanded_attention_mask.cumsum(-1) - 1).masked_fill_((expanded_attention_mask == 0), 1)
+
+    return {
+        "input_ids": expanded_input_ids,
+        "attention_mask": expanded_attention_mask,
+        "position_ids": position_ids,
+    }
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
+    **kwargs,
+):
+    r"""
+    Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+    custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+            must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+    encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+    if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+        return encoder_output.latent_dist.sample(generator)
+    elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+        return encoder_output.latent_dist.mode()
+    elif hasattr(encoder_output, "latents"):
+        return encoder_output.latents
+    else:
+        raise AttributeError("Could not access latents of provided encoder_output")
+
+
+class HunyuanVideoImageToVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
+    r"""
+    Pipeline for image-to-video generation using HunyuanVideo.
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+    implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+    Args:
+        text_encoder ([`LlavaForConditionalGeneration`]):
+            [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
+        tokenizer (`LlamaTokenizer`):
+            Tokenizer from [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
+        transformer ([`HunyuanVideoTransformer3DModel`]):
+            Conditional Transformer to denoise the encoded image latents.
+        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
+            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
+        vae ([`AutoencoderKLHunyuanVideo`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
+        text_encoder_2 ([`CLIPTextModel`]):
+            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+        tokenizer_2 (`CLIPTokenizer`):
+            Tokenizer of class
+            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
+    """
+
+    model_cpu_offload_seq = "text_encoder->text_encoder_2->transformer->vae"
+    _callback_tensor_inputs = ["latents", "prompt_embeds"]
+
+    def __init__(
+        self,
+        text_encoder: LlavaForConditionalGeneration,
+        tokenizer: LlamaTokenizerFast,
+        transformer: HunyuanVideoTransformer3DModel,
+        vae: AutoencoderKLHunyuanVideo,
+        scheduler: FlowMatchEulerDiscreteScheduler,
+        text_encoder_2: CLIPTextModel,
+        tokenizer_2: CLIPTokenizer,
+        image_processor: CLIPImageProcessor,
+    ):
+        super().__init__()
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            transformer=transformer,
+            scheduler=scheduler,
+            text_encoder_2=text_encoder_2,
+            tokenizer_2=tokenizer_2,
+            image_processor=image_processor,
+        )
+
+        self.vae_scaling_factor = self.vae.config.scaling_factor if getattr(self, "vae", None) else 0.476986
+        self.vae_scale_factor_temporal = self.vae.temporal_compression_ratio if getattr(self, "vae", None) else 4
+        self.vae_scale_factor_spatial = self.vae.spatial_compression_ratio if getattr(self, "vae", None) else 8
+        self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
+
+    def _get_llama_prompt_embeds(
+        self,
+        image: torch.Tensor,
+        prompt: Union[str, List[str]],
+        prompt_template: Dict[str, Any],
+        num_videos_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 256,
+        num_hidden_layers_to_skip: int = 2,
+        image_embed_interleave: int = 2,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        prompt = [prompt_template["template"].format(p) for p in prompt]
+
+        crop_start = prompt_template.get("crop_start", None)
+
+        image_emb_len = prompt_template.get("image_emb_len", 576)
+        image_emb_start = prompt_template.get("image_emb_start", 5)
+        image_emb_end = prompt_template.get("image_emb_end", 581)
+        double_return_token_id = prompt_template.get("double_return_token_id", 271)
+
+        if crop_start is None:
+            prompt_template_input = self.tokenizer(
+                prompt_template["template"],
+                padding="max_length",
+                return_tensors="pt",
+                return_length=False,
+                return_overflowing_tokens=False,
+                return_attention_mask=False,
+            )
+            crop_start = prompt_template_input["input_ids"].shape[-1]
+            # Remove <|start_header_id|>, <|end_header_id|>, assistant, <|eot_id|>, and placeholder {}
+            crop_start -= 5
+
+        max_sequence_length += crop_start
+        text_inputs = self.tokenizer(
+            prompt,
+            max_length=max_sequence_length,
+            padding="max_length",
+            truncation=True,
+            return_tensors="pt",
+            return_length=False,
+            return_overflowing_tokens=False,
+            return_attention_mask=True,
+        )
+        text_input_ids = text_inputs.input_ids.to(device=device)
+        prompt_attention_mask = text_inputs.attention_mask.to(device=device)
+
+        image_embeds = self.image_processor(image, return_tensors="pt").pixel_values.to(device)
+
+        image_token_index = self.text_encoder.config.image_token_index
+        pad_token_id = self.text_encoder.config.pad_token_id
+        expanded_inputs = _expand_input_ids_with_image_tokens(
+            text_input_ids,
+            prompt_attention_mask,
+            max_sequence_length,
+            image_token_index,
+            image_emb_len,
+            image_emb_start,
+            image_emb_end,
+            pad_token_id,
+        )
+        prompt_embeds = self.text_encoder(
+            **expanded_inputs,
+            pixel_values=image_embeds,
+            output_hidden_states=True,
+        ).hidden_states[-(num_hidden_layers_to_skip + 1)]
+        prompt_embeds = prompt_embeds.to(dtype=dtype)
+
+        if crop_start is not None and crop_start > 0:
+            text_crop_start = crop_start - 1 + image_emb_len
+            batch_indices, last_double_return_token_indices = torch.where(text_input_ids == double_return_token_id)
+
+            if last_double_return_token_indices.shape[0] == 3:
+                # in case the prompt is too long
+                last_double_return_token_indices = torch.cat(
+                    (last_double_return_token_indices, torch.tensor([text_input_ids.shape[-1]]))
+                )
+                batch_indices = torch.cat((batch_indices, torch.tensor([0])))
+
+            last_double_return_token_indices = last_double_return_token_indices.reshape(text_input_ids.shape[0], -1)[
+                :, -1
+            ]
+            batch_indices = batch_indices.reshape(text_input_ids.shape[0], -1)[:, -1]
+            assistant_crop_start = last_double_return_token_indices - 1 + image_emb_len - 4
+            assistant_crop_end = last_double_return_token_indices - 1 + image_emb_len
+            attention_mask_assistant_crop_start = last_double_return_token_indices - 4
+            attention_mask_assistant_crop_end = last_double_return_token_indices
+
+            prompt_embed_list = []
+            prompt_attention_mask_list = []
+            image_embed_list = []
+            image_attention_mask_list = []
+
+            for i in range(text_input_ids.shape[0]):
+                prompt_embed_list.append(
+                    torch.cat(
+                        [
+                            prompt_embeds[i, text_crop_start : assistant_crop_start[i].item()],
+                            prompt_embeds[i, assistant_crop_end[i].item() :],
+                        ]
+                    )
+                )
+                prompt_attention_mask_list.append(
+                    torch.cat(
+                        [
+                            prompt_attention_mask[i, crop_start : attention_mask_assistant_crop_start[i].item()],
+                            prompt_attention_mask[i, attention_mask_assistant_crop_end[i].item() :],
+                        ]
+                    )
+                )
+                image_embed_list.append(prompt_embeds[i, image_emb_start:image_emb_end])
+                image_attention_mask_list.append(
+                    torch.ones(image_embed_list[-1].shape[0]).to(prompt_embeds.device).to(prompt_attention_mask.dtype)
+                )
+
+            prompt_embed_list = torch.stack(prompt_embed_list)
+            prompt_attention_mask_list = torch.stack(prompt_attention_mask_list)
+            image_embed_list = torch.stack(image_embed_list)
+            image_attention_mask_list = torch.stack(image_attention_mask_list)
+
+            if 0 < image_embed_interleave < 6:
+                image_embed_list = image_embed_list[:, ::image_embed_interleave, :]
+                image_attention_mask_list = image_attention_mask_list[:, ::image_embed_interleave]
+
+            assert (
+                prompt_embed_list.shape[0] == prompt_attention_mask_list.shape[0]
+                and image_embed_list.shape[0] == image_attention_mask_list.shape[0]
+            )
+
+            prompt_embeds = torch.cat([image_embed_list, prompt_embed_list], dim=1)
+            prompt_attention_mask = torch.cat([image_attention_mask_list, prompt_attention_mask_list], dim=1)
+
+        return prompt_embeds, prompt_attention_mask
+
+    def _get_clip_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]],
+        num_videos_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 77,
+    ) -> torch.Tensor:
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder_2.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+
+        text_inputs = self.tokenizer_2(
+            prompt,
+            padding="max_length",
+            max_length=max_sequence_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+
+        text_input_ids = text_inputs.input_ids
+        untruncated_ids = self.tokenizer_2(prompt, padding="longest", return_tensors="pt").input_ids
+        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+            removed_text = self.tokenizer_2.batch_decode(untruncated_ids[:, max_sequence_length - 1 : -1])
+            logger.warning(
+                "The following part of your input was truncated because CLIP can only handle sequences up to"
+                f" {max_sequence_length} tokens: {removed_text}"
+            )
+
+        prompt_embeds = self.text_encoder_2(text_input_ids.to(device), output_hidden_states=False).pooler_output
+        return prompt_embeds
+
+    def encode_prompt(
+        self,
+        image: torch.Tensor,
+        prompt: Union[str, List[str]],
+        prompt_2: Union[str, List[str]] = None,
+        prompt_template: Dict[str, Any] = DEFAULT_PROMPT_TEMPLATE,
+        num_videos_per_prompt: int = 1,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_attention_mask: Optional[torch.Tensor] = None,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 256,
+        image_embed_interleave: int = 2,
+    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        if prompt_embeds is None:
+            prompt_embeds, prompt_attention_mask = self._get_llama_prompt_embeds(
+                image,
+                prompt,
+                prompt_template,
+                num_videos_per_prompt,
+                device=device,
+                dtype=dtype,
+                max_sequence_length=max_sequence_length,
+                image_embed_interleave=image_embed_interleave,
+            )
+
+        if pooled_prompt_embeds is None:
+            if prompt_2 is None:
+                prompt_2 = prompt
+            pooled_prompt_embeds = self._get_clip_prompt_embeds(
+                prompt,
+                num_videos_per_prompt,
+                device=device,
+                dtype=dtype,
+                max_sequence_length=77,
+            )
+
+        return prompt_embeds, pooled_prompt_embeds, prompt_attention_mask
+
+    def check_inputs(
+        self,
+        prompt,
+        prompt_2,
+        height,
+        width,
+        prompt_embeds=None,
+        callback_on_step_end_tensor_inputs=None,
+        prompt_template=None,
+        true_cfg_scale=1.0,
+        guidance_scale=1.0,
+    ):
+        if height % 16 != 0 or width % 16 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 16 but are {height} and {width}.")
+
+        if callback_on_step_end_tensor_inputs is not None and not all(
+            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+        ):
+            raise ValueError(
+                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+            )
+
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt_2 is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+        elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+            raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+        if prompt_template is not None:
+            if not isinstance(prompt_template, dict):
+                raise ValueError(f"`prompt_template` has to be of type `dict` but is {type(prompt_template)}")
+            if "template" not in prompt_template:
+                raise ValueError(
+                    f"`prompt_template` has to contain a key `template` but only found {prompt_template.keys()}"
+                )
+
+        if true_cfg_scale > 1.0 and guidance_scale > 1.0:
+            logger.warning(
+                "Both `true_cfg_scale` and `guidance_scale` are greater than 1.0. This will result in both "
+                "classifier-free guidance and embedded-guidance to be applied. This is not recommended "
+                "as it may lead to higher memory usage, slower inference and potentially worse results."
+            )
+
+    def prepare_latents(
+        self,
+        image: torch.Tensor,
+        batch_size: int,
+        num_channels_latents: int = 32,
+        height: int = 720,
+        width: int = 1280,
+        num_frames: int = 129,
+        dtype: Optional[torch.dtype] = None,
+        device: Optional[torch.device] = None,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+        image_condition_type: str = "latent_concat",
+    ) -> torch.Tensor:
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        num_latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
+        latent_height, latent_width = height // self.vae_scale_factor_spatial, width // self.vae_scale_factor_spatial
+        shape = (batch_size, num_channels_latents, num_latent_frames, latent_height, latent_width)
+
+        image = image.unsqueeze(2)  # [B, C, 1, H, W]
+        if isinstance(generator, list):
+            image_latents = [
+                retrieve_latents(self.vae.encode(image[i].unsqueeze(0)), generator[i], "argmax")
+                for i in range(batch_size)
+            ]
+        else:
+            image_latents = [retrieve_latents(self.vae.encode(img.unsqueeze(0)), generator, "argmax") for img in image]
+
+        image_latents = torch.cat(image_latents, dim=0).to(dtype) * self.vae_scaling_factor
+        image_latents = image_latents.repeat(1, 1, num_latent_frames, 1, 1)
+
+        if latents is None:
+            latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+        else:
+            latents = latents.to(device=device, dtype=dtype)
+
+        t = torch.tensor([0.999]).to(device=device)
+        latents = latents * t + image_latents * (1 - t)
+
+        if image_condition_type == "token_replace":
+            image_latents = image_latents[:, :, :1]
+
+        return latents, image_latents
+
+    def enable_vae_slicing(self):
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.vae.enable_slicing()
+
+    def disable_vae_slicing(self):
+        r"""
+        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_slicing()
+
+    def enable_vae_tiling(self):
+        r"""
+        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+        processing larger images.
+        """
+        self.vae.enable_tiling()
+
+    def disable_vae_tiling(self):
+        r"""
+        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_tiling()
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    @property
+    def num_timesteps(self):
+        return self._num_timesteps
+
+    @property
+    def attention_kwargs(self):
+        return self._attention_kwargs
+
+    @property
+    def current_timestep(self):
+        return self._current_timestep
+
+    @property
+    def interrupt(self):
+        return self._interrupt
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        image: PIL.Image.Image,
+        prompt: Union[str, List[str]] = None,
+        prompt_2: Union[str, List[str]] = None,
+        negative_prompt: Union[str, List[str]] = None,
+        negative_prompt_2: Union[str, List[str]] = None,
+        height: int = 720,
+        width: int = 1280,
+        num_frames: int = 129,
+        num_inference_steps: int = 50,
+        sigmas: List[float] = None,
+        true_cfg_scale: float = 1.0,
+        guidance_scale: float = 1.0,
+        num_videos_per_prompt: Optional[int] = 1,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_attention_mask: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        attention_kwargs: Optional[Dict[str, Any]] = None,
+        callback_on_step_end: Optional[
+            Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+        ] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        prompt_template: Dict[str, Any] = DEFAULT_PROMPT_TEMPLATE,
+        max_sequence_length: int = 256,
+        image_embed_interleave: Optional[int] = None,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+                will be used instead.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
+                not greater than `1`).
+            negative_prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+                `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
+            height (`int`, defaults to `720`):
+                The height in pixels of the generated image.
+            width (`int`, defaults to `1280`):
+                The width in pixels of the generated image.
+            num_frames (`int`, defaults to `129`):
+                The number of frames in the generated video.
+            num_inference_steps (`int`, defaults to `50`):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
+            true_cfg_scale (`float`, *optional*, defaults to 1.0):
+                When > 1.0 and a provided `negative_prompt`, enables true classifier-free guidance.
+            guidance_scale (`float`, defaults to `1.0`):
+                Guidance scale as defined in [Classifier-Free Diffusion
+                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+                the text `prompt`, usually at the expense of lower image quality. Note that the only available
+                HunyuanVideo model is CFG-distilled, which means that traditional guidance between unconditional and
+                conditional latent is not applied.
+            num_videos_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            latents (`torch.Tensor`, *optional*):
+                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor is generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.Tensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+                input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`HunyuanVideoPipelineOutput`] instead of a plain tuple.
+            attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
+            callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+                A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+                each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+                DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+                list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeline class.
+
+        Examples:
+
+        Returns:
+            [`~HunyuanVideoPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`HunyuanVideoPipelineOutput`] is returned, otherwise a `tuple` is returned
+                where the first element is a list with the generated images and the second element is a list of `bool`s
+                indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw) content.
+        """
+
+        if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+            callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            prompt_2,
+            height,
+            width,
+            prompt_embeds,
+            callback_on_step_end_tensor_inputs,
+            prompt_template,
+            true_cfg_scale,
+            guidance_scale,
+        )
+
+        image_condition_type = self.transformer.config.image_condition_type
+        has_neg_prompt = negative_prompt is not None or (
+            negative_prompt_embeds is not None and negative_pooled_prompt_embeds is not None
+        )
+        do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
+        image_embed_interleave = (
+            image_embed_interleave
+            if image_embed_interleave is not None
+            else (
+                2 if image_condition_type == "latent_concat" else 4 if image_condition_type == "token_replace" else 1
+            )
+        )
+
+        self._guidance_scale = guidance_scale
+        self._attention_kwargs = attention_kwargs
+        self._current_timestep = None
+        self._interrupt = False
+
+        device = self._execution_device
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        # 3. Prepare latent variables
+        vae_dtype = self.vae.dtype
+        image_tensor = self.video_processor.preprocess(image, height, width).to(device, vae_dtype)
+
+        if image_condition_type == "latent_concat":
+            num_channels_latents = (self.transformer.config.in_channels - 1) // 2
+        elif image_condition_type == "token_replace":
+            num_channels_latents = self.transformer.config.in_channels
+
+        latents, image_latents = self.prepare_latents(
+            image_tensor,
+            batch_size * num_videos_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            num_frames,
+            torch.float32,
+            device,
+            generator,
+            latents,
+            image_condition_type,
+        )
+        if image_condition_type == "latent_concat":
+            image_latents[:, :, 1:] = 0
+            mask = image_latents.new_ones(image_latents.shape[0], 1, *image_latents.shape[2:])
+            mask[:, :, 1:] = 0
+
+        # 4. Encode input prompt
+        transformer_dtype = self.transformer.dtype
+        prompt_embeds, pooled_prompt_embeds, prompt_attention_mask = self.encode_prompt(
+            image=image,
+            prompt=prompt,
+            prompt_2=prompt_2,
+            prompt_template=prompt_template,
+            num_videos_per_prompt=num_videos_per_prompt,
+            prompt_embeds=prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            prompt_attention_mask=prompt_attention_mask,
+            device=device,
+            max_sequence_length=max_sequence_length,
+            image_embed_interleave=image_embed_interleave,
+        )
+        prompt_embeds = prompt_embeds.to(transformer_dtype)
+        prompt_attention_mask = prompt_attention_mask.to(transformer_dtype)
+        pooled_prompt_embeds = pooled_prompt_embeds.to(transformer_dtype)
+
+        if do_true_cfg:
+            black_image = PIL.Image.new("RGB", (width, height), 0)
+            negative_prompt_embeds, negative_pooled_prompt_embeds, negative_prompt_attention_mask = self.encode_prompt(
+                image=black_image,
+                prompt=negative_prompt,
+                prompt_2=negative_prompt_2,
+                prompt_template=prompt_template,
+                num_videos_per_prompt=num_videos_per_prompt,
+                prompt_embeds=negative_prompt_embeds,
+                pooled_prompt_embeds=negative_pooled_prompt_embeds,
+                prompt_attention_mask=negative_prompt_attention_mask,
+                device=device,
+                max_sequence_length=max_sequence_length,
+            )
+            negative_prompt_embeds = negative_prompt_embeds.to(transformer_dtype)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.to(transformer_dtype)
+            negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.to(transformer_dtype)
+
+        # 5. Prepare timesteps
+        sigmas = np.linspace(1.0, 0.0, num_inference_steps + 1)[:-1] if sigmas is None else sigmas
+        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, sigmas=sigmas)
+
+        # 6. Prepare guidance condition
+        guidance = None
+        if self.transformer.config.guidance_embeds:
+            guidance = (
+                torch.tensor([guidance_scale] * latents.shape[0], dtype=transformer_dtype, device=device) * 1000.0
+            )
+
+        # 7. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        self._num_timesteps = len(timesteps)
+
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if self.interrupt:
+                    continue
+
+                self._current_timestep = t
+                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+                timestep = t.expand(latents.shape[0]).to(latents.dtype)
+
+                if image_condition_type == "latent_concat":
+                    latent_model_input = torch.cat([latents, image_latents, mask], dim=1).to(transformer_dtype)
+                elif image_condition_type == "token_replace":
+                    latent_model_input = torch.cat([image_latents, latents[:, :, 1:]], dim=2).to(transformer_dtype)
+
+                noise_pred = self.transformer(
+                    hidden_states=latent_model_input,
+                    timestep=timestep,
+                    encoder_hidden_states=prompt_embeds,
+                    encoder_attention_mask=prompt_attention_mask,
+                    pooled_projections=pooled_prompt_embeds,
+                    guidance=guidance,
+                    attention_kwargs=attention_kwargs,
+                    return_dict=False,
+                )[0]
+
+                if do_true_cfg:
+                    neg_noise_pred = self.transformer(
+                        hidden_states=latent_model_input,
+                        timestep=timestep,
+                        encoder_hidden_states=negative_prompt_embeds,
+                        encoder_attention_mask=negative_prompt_attention_mask,
+                        pooled_projections=negative_pooled_prompt_embeds,
+                        guidance=guidance,
+                        attention_kwargs=attention_kwargs,
+                        return_dict=False,
+                    )[0]
+                    noise_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)
+
+                # compute the previous noisy sample x_t -> x_t-1
+                if image_condition_type == "latent_concat":
+                    latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+                elif image_condition_type == "token_replace":
+                    latents = latents = self.scheduler.step(
+                        noise_pred[:, :, 1:], t, latents[:, :, 1:], return_dict=False
+                    )[0]
+                    latents = torch.cat([image_latents, latents], dim=2)
+
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
+        self._current_timestep = None
+
+        if not output_type == "latent":
+            latents = latents.to(self.vae.dtype) / self.vae_scaling_factor
+            video = self.vae.decode(latents, return_dict=False)[0]
+            if image_condition_type == "latent_concat":
+                video = video[:, :, 4:, :, :]
+            video = self.video_processor.postprocess_video(video, output_type=output_type)
+        else:
+            if image_condition_type == "latent_concat":
+                video = latents[:, :, 1:, :, :]
+            else:
+                video = latents
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (video,)
+
+        return HunyuanVideoPipelineOutput(frames=video)