Instructions to use BingoG/SpatialAV2AV with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LTX.io
How to use BingoG/SpatialAV2AV with LTX.io:
# Install the LTX-2 pipelines git clone https://github.com/Lightricks/LTX-2.git cd LTX-2 uv sync --frozen
# Download the weights from this repo, plus the Gemma text encoder hf download BingoG/SpatialAV2AV --local-dir models/SpatialAV2AV hf download google/gemma-3-12b-it-qat-q4_0-unquantized --local-dir models/gemma-3-12b
# Fast pipeline (distilled model, no distilled LoRA needed) uv run python -m ltx_pipelines.distilled \ --distilled-checkpoint-path models/SpatialAV2AV/<distilled-checkpoint>.safetensors \ --spatial-upsampler-path models/SpatialAV2AV/<spatial-upsampler>.safetensors \ --gemma-root models/gemma-3-12b \ --prompt "A beautiful sunset over the ocean" \ --output-path output.mp4 # For image-to-video, add: --image path/to/image.jpg 0 0.8# HQ pipeline (two-stage, higher quality) uv run python -m ltx_pipelines.ti2vid_two_stages_hq \ --checkpoint-path models/SpatialAV2AV/<checkpoint>.safetensors \ --distilled-lora models/SpatialAV2AV/<distilled-lora>.safetensors 0.8 \ --spatial-upsampler-path models/SpatialAV2AV/<spatial-upsampler>.safetensors \ --gemma-root models/gemma-3-12b \ --prompt "A beautiful sunset over the ocean" \ --output-path output.mp4 # For image-to-video, add: --image path/to/image.jpg 0 0.8 - Notebooks
- Google Colab
- Kaggle
| # test_set_media_1000 — self-contained media for 1000 test samples | |
| Full media for the **first 1,000 test samples** (pan/rotate trajectories) of the | |
| SpatialAV2AV held-out test split, so the test set is directly usable **without** the | |
| `BingoG/LTX` dataset repo. | |
| ## Why this exists | |
| The test manifests in [`test_set/`](../test_set) list 11,616 samples drawn from the 116k | |
| `all.list` split. The dataset repo `BingoG/LTX` only carries a 4,400-sample debug subset, | |
| so it does **not** cover this test set. This folder ships real, self-contained media for a | |
| usable 1,000-sample slice (pan/rotate — the trajectories the I3 counterfactual evaluation | |
| needs). | |
| ## Contents (4,251 files, ~2.24 GB) | |
| | Folder | Count | What | | |
| |--------|------:|------| | |
| | `final_json/` | 1000 | metadata JSON per sample (relative media paths inside) | | |
| | `final_source/` | 1000 | source (condition) video, mp4 with embedded stereo | | |
| | `final_edit/` | 1000 | edit (target) video, mp4 with embedded stereo | | |
| | `final_audio_edit/` | 1000 | target stereo audio, wav (used for tILD / rank GT) | | |
| | `final_audio_src/` | 250 | source stereo audio, wav (shared per origin) | | |
| | `samples_1000.relative.list` | 1 | the 1,000 sample json paths (relative) | | |
| Layout mirrors the `speakervid` dataset root, so each json's relative paths | |
| (`final_source/<clip>.mp4`, etc.) resolve against this folder directly. | |
| ## Usage | |
| ```python | |
| from huggingface_hub import snapshot_download | |
| root = snapshot_download("BingoG/SpatialAV2AV", allow_patterns="test_set_media_1000/*") | |
| base = f"{root}/test_set_media_1000" | |
| # each final_json/<clip>.json has relative paths (final_source/…, final_edit/…, | |
| # final_audio_edit/…) that resolve under `base` — point the evaluator there. | |
| ``` | |
| ## Scope note | |
| - These 1,000 are a **subset** of the full 11,616-sample test set (`test_set/test.list`). | |
| They are all **pan/rotate** samples. The remaining test samples (other pan/rotate + | |
| fixed/push_in/pull_out) are not included here yet. | |
| - Media is real 2-channel/44.1 kHz stereo embedded in the mp4s, plus explicit stereo wavs. | |