Upload 72 files

.gitattributes

@@ -1,35 +1,39 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.webp filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,149 @@
+debug/
+testing/output/
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# PyCharm
+.idea
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# MacOS filesystem
+.DS_Store
+
+.gradio

README.md

@@ -1,12 +1,29 @@
----
-title: ISSATM VTO
-emoji: 👁
-colorFrom: red
-colorTo: gray
-sdk: gradio
-sdk_version: 5.6.0
-app_file: app.py
-pinned: false
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+---
+title: Virtual Try-On Diffusion [VTON-D]
+emoji: 👗
+colorFrom: indigo
+colorTo: purple
+sdk: gradio
+sdk_version: 5.6.0
+app_file: app.py
+pinned: false
+short_description: Diffusion-based multi-modal virtual try-on pipeline demo
+tags:
+    - virtual try-on
+    - vton
+    - clothing transfer
+    - diffusion
+    - img2img
+    - txt2img
+---
+
+# Virtual Try-On Diffusion [VTON-D] by Texel.Moda
+
+Virtual Try-On Diffusion [VTON-D] by [Texel.Moda](https://texelmoda.com/) is a custom diffusion-based pipeline for fast 
+and flexible multi-modal virtual try-on. Clothing, avatar and background can be specified by reference images or text 
+prompts allowing for clothing transfer, avatar replacement, fashion image generation and other virtual try-on related 
+tasks.
+
+This API is available [through the RapidAPI Hub](https://rapidapi.com/texelmoda-texelmoda-apis/api/try-on-diffusion).
+Additionally, you can find the extensive API documentation 
+[here](https://huggingface.co/spaces/texelmoda/try-on-diffusion/blob/main/docs/API.md).

app.py

@@ -0,0 +1,291 @@
+import logging
+import gradio as gr
+import numpy as np
+import cv2
+import os
+import base64
+
+from try_on_diffusion_client import TryOnDiffusionClient
+
+LOG_LEVEL = logging.INFO
+LOG_FORMAT = "%(asctime)s %(thread)-8s %(name)-16s %(levelname)-8s %(message)s"
+LOG_DATE_FORMAT = "%Y-%m-%d %H:%M:%S"
+
+EXAMPLE_PATH = os.path.join(os.path.dirname(__file__), "examples")
+
+API_URL = os.getenv("TRY_ON_DIFFUSION_DEMO_API_URL", "http://localhost:8000")
+API_KEY = os.getenv("TRY_ON_DIFFUSION_DEMO_API_KEY", "")
+
+SHOW_RAPIDAPI_LINK = os.getenv("TRY_ON_DIFFUSION_DEMO_SHOW_RAPIDAPI_LINK", "1") == "1"
+
+CONCURRENCY_LIMIT = int(os.getenv("TRY_ON_DIFFUSION_DEMO_CONCURRENCY_LIMIT", "2"))
+
+logging.basicConfig(level=LOG_LEVEL, format=LOG_FORMAT, datefmt=LOG_DATE_FORMAT)
+
+client = TryOnDiffusionClient(base_url=API_URL, api_key=API_KEY)
+
+
+def get_image_base64(file_name: str) -> str:
+    _, ext = os.path.splitext(file_name.lower())
+
+    content_type = "image/jpeg"
+
+    if ext == ".png":
+        content_type = "image/png"
+    elif ext == ".webp":
+        content_type = "image/webp"
+    elif ext == ".gif":
+        content_type = "image/gif"
+
+    with open(file_name, "rb") as f:
+        return f"data:{content_type};base64," + base64.b64encode(f.read()).decode("utf-8")
+
+
+def get_examples(example_dir: str) -> list[str]:
+    file_list = [f for f in os.listdir(os.path.join(EXAMPLE_PATH, example_dir)) if f.endswith(".jpg")]
+    file_list.sort()
+
+    return [os.path.join(EXAMPLE_PATH, example_dir, f) for f in file_list]
+
+
+def try_on(
+    clothing_image: np.ndarray = None,
+    clothing_prompt: str = None,
+    avatar_image: np.ndarray = None,
+    avatar_prompt: str = None,
+    avatar_sex: str = None,
+    background_image: np.ndarray = None,
+    background_prompt: str = None,
+    seed: int = -1,
+) -> tuple:
+    result = client.try_on_file(
+        clothing_image=cv2.cvtColor(clothing_image, cv2.COLOR_RGB2BGR) if clothing_image is not None else None,
+        clothing_prompt=clothing_prompt,
+        avatar_image=cv2.cvtColor(avatar_image, cv2.COLOR_RGB2BGR) if avatar_image is not None else None,
+        avatar_prompt=avatar_prompt,
+        avatar_sex=avatar_sex if avatar_sex in ["male", "female"] else None,
+        background_image=cv2.cvtColor(background_image, cv2.COLOR_RGB2BGR) if background_image is not None else None,
+        background_prompt=background_prompt,
+        seed=seed,
+    )
+
+    if result.status_code == 200:
+        return cv2.cvtColor(result.image, cv2.COLOR_BGR2RGB), f"<h3>Success</h3><p>Seed: {result.seed}</p>"
+    else:
+        error_message = f"<h3>Error {result.status_code}</h3>"
+
+        if result.error_details is not None:
+            error_message += f"<p>{result.error_details}</p>"
+
+        return None, error_message
+
+
+with gr.Blocks(theme=gr.themes.Soft(), delete_cache=(3600, 3600)) as app:
+    gr.HTML(
+        f"""
+        <div style="width: 100%; background-color: #001537; border-radius: 10px; padding-left: 10px">
+            <a href="https://texelmoda.com/" target="_blank">
+                <img src="{get_image_base64("images/logo.png")}" title="Texel.Moda" alt="Texel.Moda" style="float: left; margin-right: 10px; margin-left: -10px; border-radius: 10px; max-height: 50px;"/>
+            </a>
+            <h1 style="margin: 0; margin-right: 10px; line-height: 50px; color: #8cecd5; text-transform: uppercase">Virtual Try-On Diffusion</h1>
+        </div>
+        <br/>
+        <p>
+            Virtual Try-On Diffusion [VTON-D] by <a href="https://texelmoda.com/" target="_blank">Texel.Moda</a> is a 
+            custom diffusion-based pipeline for fast and flexible multi-modal virtual try-on.
+            Clothing, avatar and background can be specified by reference images or text prompts allowing for clothing 
+            transfer, avatar replacement, fashion image generation and other virtual try-on related tasks. 
+    """
+    )
+
+    if SHOW_RAPIDAPI_LINK:
+        gr.Button(
+            value="Check out the API @ RapidAPI.com",
+            link="https://rapidapi.com/texelmoda-texelmoda-apis/api/try-on-diffusion",
+            icon="https://files.readme.io/9336831-small-rapid-logo-favicon.png",
+        )
+
+        gr.HTML(
+            """
+            <center>
+                <a href="https://huggingface.co/spaces/texelmoda/try-on-diffusion/blob/main/docs/API.md">API Documentation</a>
+            </center>
+            """
+        )
+
+    gr.HTML("</p>")
+
+    with gr.Row():
+        with gr.Column():
+            gr.HTML(
+                """
+                <h2>Clothing</h2>
+                <p>
+                    Clothing may be specified with a reference image or a text prompt. 
+                    For more exotic use cases image and prompt can be also used together.
+                    If both image and prompt are empty the model will generate random clothing.
+                    <br/><br/>
+                </p>
+                """
+            )
+
+            with gr.Tab("Image"):
+                clothing_image = gr.Image(label="Clothing Image", sources=["upload"], type="numpy")
+
+                clothing_image_examples = gr.Examples(
+                    inputs=clothing_image, examples_per_page=12, examples=get_examples("clothing")
+                )
+
+            with gr.Tab("Prompt"):
+                clothing_prompt = gr.TextArea(
+                    label="Clothing Prompt",
+                    info='Compel weighting <a href="https://github.com/damian0815/compel/blob/main/doc/syntax.md">syntax</a> is supported.',
+                )
+
+                clothing_prompt_examples = gr.Examples(
+                    inputs=clothing_prompt,
+                    examples_per_page=8,
+                    examples=[
+                        "a sheer blue sleeveless mini dress",
+                        "a beige woolen sweater and white pleated skirt",
+                        "a black leather jacket and dark blue slim-fit jeans",
+                        "a floral pattern blouse and leggings",
+                        "a paisley pattern purple shirt and beige chinos",
+                        "a striped white and blue polo shirt and blue jeans",
+                        "a colorful t-shirt and black shorts",
+                        "a checked pattern shirt and dark blue cargo pants",
+                    ],
+                )
+
+        with gr.Column():
+            gr.HTML(
+                """
+                <h2>Avatar</h2>
+                <p>
+                    Avatar may be specified with a subject photo or a text prompt.
+                    Latter can be used, for example, to replace person while preserving clothing. 
+                    For more exotic use cases image and prompt can be also used together.
+                    If both image and prompt are empty the model will generate random avatars.
+                </p>
+                """
+            )
+
+            with gr.Tab("Image"):
+                avatar_image = gr.Image(label="Avatar Image", sources=["upload"], type="numpy")
+
+                avatar_image_examples = gr.Examples(
+                    inputs=avatar_image,
+                    examples_per_page=12,
+                    examples=get_examples("avatar"),
+                )
+
+            with gr.Tab("Prompt"):
+                avatar_prompt = gr.TextArea(
+                    label="Avatar Prompt",
+                    info='Compel weighting <a href="https://github.com/damian0815/compel/blob/main/doc/syntax.md">syntax</a> is supported.',
+                )
+
+                avatar_prompt_examples = gr.Examples(
+                    inputs=avatar_prompt,
+                    examples_per_page=8,
+                    examples=[
+                        "a beautiful blond girl with long hair",
+                        "a cute redhead girl with freckles",
+                        "a plus size female model wearing sunglasses",
+                        "a woman with dark hair and blue eyes",
+                        "a fit man with dark beard and blue eyes",
+                        "a young blond man posing for a photo",
+                        "a gentleman with beard and mustache",
+                        "a plus size man walking",
+                    ],
+                )
+
+            avatar_sex = gr.Dropdown(
+                label="Avatar Sex",
+                choices=[("Auto", ""), ("Male", "male"), ("Female", "female")],
+                value="",
+                info="Avatar sex selector can be used to enforce a specific sex of the avatar.",
+            )
+
+        with gr.Column():
+            gr.HTML(
+                """
+                <h2>Background</h2>
+                <p>
+                    Replacing background is optional. 
+                    Background may be specified with a reference image or a text prompt.
+                    If omitted original avatar background will be preserved.
+                    <br/><br/><br/>
+                </p>
+                """
+            )
+
+            with gr.Tab("Image"):
+                background_image = gr.Image(label="Background Image", sources=["upload"], type="numpy")
+
+                background_image_examples = gr.Examples(
+                    inputs=background_image, examples_per_page=12, examples=get_examples("background")
+                )
+
+            with gr.Tab("Prompt"):
+                background_prompt = gr.TextArea(
+                    label="Background Prompt",
+                    info='Compel weighting <a href="https://github.com/damian0815/compel/blob/main/doc/syntax.md">syntax</a> is supported.',
+                )
+
+                background_prompt_examples = gr.Examples(
+                    inputs=background_prompt,
+                    examples_per_page=8,
+                    examples=[
+                        "in an autumn park",
+                        "in front of a brick wall",
+                        "near an old tree",
+                        "on a busy city street",
+                        "in front of a staircase",
+                        "on an ocean beach with palm trees",
+                        "in a shopping mall",
+                        "in a modern office",
+                    ],
+                )
+
+        with gr.Column():
+            gr.HTML(
+                """
+                <h2>Generation</h2>
+                """
+            )
+
+            seed = gr.Number(
+                label="Seed",
+                value=-1,
+                minimum=-1,
+                info="Seed used for generation, specify -1 for random seed for each generation.",
+            )
+
+            generate_button = gr.Button(value="Generate", variant="primary")
+
+            result_image = gr.Image(label="Result", show_share_button=False, format="jpeg")
+            result_details = gr.HTML(label="Details")
+
+    generate_button.click(
+        fn=try_on,
+        inputs=[
+            clothing_image,
+            clothing_prompt,
+            avatar_image,
+            avatar_prompt,
+            avatar_sex,
+            background_image,
+            background_prompt,
+            seed,
+        ],
+        outputs=[result_image, result_details],
+        api_name=False,
+        concurrency_limit=CONCURRENCY_LIMIT,
+    )
+
+    app.title = "Virtual Try-On Diffusion by Texel.Moda"
+
+
+if __name__ == "__main__":
+    app.queue(api_open=False).launch(show_api=False)

docs/API.md

@@ -0,0 +1,341 @@
+# Virtual Try-On Diffusion API
+
+<!-- TOC -->
+* [Virtual Try-On Diffusion API](#virtual-try-on-diffusion-api)
+  * [Summary](#summary)
+  * [Consuming the API](#consuming-the-api)
+  * [Try-On Endpoints](#try-on-endpoints)
+  * [Try-On Input Parameters](#try-on-input-parameters)
+    * [Clothing image](#clothing-image)
+    * [Clothing prompt](#clothing-prompt)
+    * [Avatar image](#avatar-image)
+    * [Avatar prompt](#avatar-prompt)
+    * [Background image](#background-image)
+    * [Background prompt](#background-prompt)
+    * [Additional notes](#additional-notes)
+  * [Try-On Output](#try-on-output)
+    * [Response codes](#response-codes)
+    * [NSFW content](#nsfw-content)
+  * [Use Cases and Recipes](#use-cases-and-recipes)
+    * [Image-based virtual try-on](#image-based-virtual-try-on)
+    * [Image-based virtual try-on with background](#image-based-virtual-try-on-with-background)
+    * [Avatar from a text prompt](#avatar-from-a-text-prompt)
+    * [Clothing from a text prompt](#clothing-from-a-text-prompt)
+    * [Modifying avatar's body](#modifying-avatars-body)
+    * [Txt2Img](#txt2img)
+    * [Other creative possibilities](#other-creative-possibilities)
+  * [Performance](#performance)
+  * [Known Issues and Limitations](#known-issues-and-limitations)
+<!-- TOC -->
+
+## Summary
+
+Virtual Try-On Diffusion [VTON-D] by [Texel.Moda](https://texelmoda.com) is a custom diffusion-based pipeline for fast 
+and flexible multi-modal virtual try-on. Clothing, avatar and background can be specified by reference images or text 
+prompts allowing for clothing transfer, avatar replacement, fashion image generation and other virtual try-on related 
+tasks. Check out the [demo on Hugging Face](https://huggingface.co/spaces/texelmoda/try-on-diffusion) to try the API in
+a user-friendly way.
+
+## Consuming the API
+
+The API is exposed through the RapidAPI Hub which manages API subscriptions, API keys, payments and other things. Please 
+refer to the [RapidAPI Documentation](https://docs.rapidapi.com/docs/consumer-quick-start-guide) to get started.
+
+Generally, in order to use an API you need to perform the following steps:
+- Create a RapidAPI.com account.
+- [Navigate to the API page](https://rapidapi.com/texelmoda-texelmoda-apis/api/try-on-diffusion) and subscribe to a 
+  suitable pricing plan. We also provide a free BASIC plan with 100 API requests per month.
+- Use the obtained RapidAPI key to authenticate (via the _X-RapidAPI-Key_ header) and use an API from any programming 
+  language or tool you like.
+
+Example API call using cURL:
+```shell
+curl --request POST \
+--url https://try-on-diffusion.p.rapidapi.com/try-on-file \
+--header 'Content-Type: multipart/form-data' \
+--header 'x-rapidapi-host: try-on-diffusion.p.rapidapi.com' \
+--header 'x-rapidapi-key: <RapidAPI Key>' \
+--form clothing_image=1.jpg \
+--form avatar_image=2.jpg
+```
+
+For a simple Python client implementation please see the 
+[Hugging Face demo application source](https://huggingface.co/spaces/texelmoda/try-on-diffusion/blob/main/try_on_diffusion_client.py).
+
+## Try-On Endpoints
+
+Try-On API consists of two endpoints that differ only in the method of passing reference images:
+
+- **POST** _/try-on-file_ - takes reference images as uploaded files in the request body (using multipart/form-data).
+
+
+- **POST** _/try-on-url_ - takes reference images as image URLs in POST parameters.
+
+All image requirements, behavior and status codes are the same for both endpoints, choose the one that best suits your 
+application architecture.
+
+## Try-On Input Parameters
+
+All input parameters for the try-on endpoints are currently optional. Images and prompts serve as additional generation 
+conditions and can even be used in combination. Below is the short parameter summary with links to extended information 
+on certain parameters.
+
+List of input parameters for the **POST** _/try-on-file_ endpoint:
+
+| Parameter                               | Description                                                                                                                                                                          | Required |
+|-----------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|----------|
+| [clothing_image](#clothing-image)       | Clothing reference image in JPEG, PNG or WEBP format, maximum file size is 12 MB.                                                                                                    | No       |
+| [clothing_prompt](#clothing-prompt)     | Text prompt for clothing, can be used instead of an image. Compel weighting syntax is supported. Example: _red sleeveless mini dress_                                                | No       |
+| [avatar_image](#avatar-image)           | Avatar image in JPEG, PNG or WEBP format, maximum file size is 12 MB.                                                                                                                | No       |
+| avatar_sex                              | Avatar sex, either "male" or "female". Will be detected automatically, if left empty or omitted. Will enforce certain avatar sex if specified.                                       | No       |
+| [avatar_prompt](#avatar-prompt)         | Text prompt for the avatar, can be used instead of an image or with image to modify the avatar. Compel weighting syntax is supported. Example: _a gentleman with beard and mustache_ | No       |
+| [background_image](#background-image)   | Optional background reference image in JPEG, PNG or WEBP format, maximum file size is 12 MB. Original avatar background is preserved if background is not specified.                 | No       |
+| [background_prompt](#background-prompt) | Optional background text prompt. Original avatar background is preserved if background is not specified. Example: _in an autumn park_                                                | No       |
+| seed                                    | Seed for image generation. Default is -1 (random seed). Actual seed will also be output in the "X-Seed" response header. Example: _42_                                               | No       |
+
+List of input parameters for the **POST** _/try-on-url_ endpoint:
+
+| Parameter                                 | Description                                                                                                                                                                               | Required |
+|-------------------------------------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|----------|
+| [clothing_image_url](#clothing-image)     | Clothing reference image URL. Image should be in JPEG, PNG or WEBP format, maximum file size is 12 MB.                                                                                    | No       |
+| [clothing_prompt](#clothing-prompt)       | Text prompt for clothing, can be used instead of an image. Compel weighting syntax is supported. Example: _red sleeveless mini dress_                                                     | No       |
+| [avatar_image_url](#avatar-image)         | Avatar image URL. Image should be in JPEG, PNG or WEBP format, maximum file size is 12 MB.                                                                                                | No       |
+| avatar_sex                                | Avatar sex, either "male" or "female". Will be detected automatically, if left empty or omitted. Will enforce certain avatar sex if specified.                                            | No       |
+| [avatar_prompt](#avatar-prompt)           | Text prompt for the avatar, can be used instead of an image or with image to modify the avatar. Compel weighting syntax is supported. Example: _a gentleman with beard and mustache_      | No       |
+| [background_image_url](#background-image) | Optional background reference image URL. Image should be in JPEG, PNG or WEBP format, maximum file size is 12 MB. Original avatar background is preserved if background is not specified. | No       |
+| [background_prompt](#background-prompt)   | Optional background text prompt. Original avatar background is preserved if background is not specified. Example: _in an autumn park_                                                     | No       |
+| seed                                      | Seed for image generation. Default is -1 (random seed). Actual seed will also be output in the "X-Seed" response header. Example: _42_                                                    | No       |
+
+### Clothing image
+
+For best results clothing reference images should meet a number of requirements:
+
+- File format: **JPEG**, **PNG** or **WEBP**
+- Maximum file size: **12 MB**
+- Minimum image size: **256x256**
+- Recommended image size: **768x1024 and above**
+- Clothing should be **dressed on a person**. Some flat lay clothing photos might work, but currently it's not guaranteed
+- **Single person** on the image (though multiple persons might also work)
+- **Frontal** photo, though some degree of rotation is fine
+- **Good lighting** conditions and **high image quality** as it directly affects the result
+- **Minimal occlusion** by hair, hands or accessories
+
+To summarize: the better is the clothing image the better is the final result.
+
+Examples of good clothing images:
+
+| <img src="images/clothing_image_01.jpg" width="240"> | <img src="images/clothing_image_02.jpg" width="240"> | <img src="images/clothing_image_03.jpg" width="240"> | <img src="images/clothing_image_04.jpg" width="240"> |
+|------------------------------------------------------|------------------------------------------------------|------------------------------------------------------|------------------------------------------------------|
+
+### Clothing prompt
+
+Instead of a clothing image you can use text prompt to describe the garment. Short and clear prompts work best. 
+Additionally, [Compel weighting syntax](https://github.com/damian0815/compel/blob/main/doc/syntax.md) is supported to 
+increase or decrease weight of certain tokens. Examples:
+- _a sheer blue sleeveless mini dress_
+- _a beige woolen sweater and white pleated skirt_
+- _a black leather jacket and dark blue slim-fit jeans_
+- _a floral pattern blouse and leggings_
+- _a colorful+++ t-shirt and black shorts_
+
+### Avatar image
+
+Avatar images should also meet a some requirements:
+
+- File format: **JPEG**, **PNG** or **WEBP**
+- Maximum file size: **12 MB**
+- Minimum image size: **256x256**
+- Recommended image size: **768x1024 and above**
+- **Single person** on the image (though multiple persons might also work)
+- **Frontal** photo, though some degree of rotation is fine
+- **Good lighting** conditions and **high image quality**
+
+Examples of good avatar images:
+
+| <img src="images/avatar_image_01.jpg" width="240"> | <img src="images/avatar_image_02.jpg" width="240"> | <img src="images/avatar_image_03.jpg" width="240"> | <img src="images/avatar_image_04.jpg" width="240"> |
+|----------------------------------------------------|----------------------------------------------------|----------------------------------------------------|----------------------------------------------------|
+
+### Avatar prompt
+
+Instead of an avatar image you can use text prompt to describe the person. Short and clear prompts work best. 
+Additionally, [Compel weighting syntax](https://github.com/damian0815/compel/blob/main/doc/syntax.md) is supported to 
+increase or decrease weight of certain tokens. Examples:
+- _a beautiful blond girl with long hair_
+- _a cute redhead girl with freckles_
+- _a (plus size)++ female model wearing sunglasses_
+- _a fit man with dark beard and blue eyes_
+- _a gentleman with beard and mustache_
+
+### Background image
+
+Background images are used to extract high-level background features only and serve as a reference (and not exact 
+background). Below are basic image requirements:
+
+- File format: **JPEG**, **PNG** or **WEBP**
+- Maximum file size: **12 MB**
+- Recommended image size: **256x256 and above**
+
+Examples of background images:
+
+| <img src="images/background_image_01.jpg" width="240"> | <img src="images/background_image_02.jpg" width="240"> | <img src="images/background_image_03.jpg" width="240"> | <img src="images/background_image_04.jpg" width="240"> |
+|--------------------------------------------------------|--------------------------------------------------------|--------------------------------------------------------|--------------------------------------------------------|
+
+### Background prompt
+
+Instead of a background image you can use text prompt to describe the background. Short and clear prompts work best. 
+Additionally, [Compel weighting syntax](https://github.com/damian0815/compel/blob/main/doc/syntax.md) is supported to 
+increase or decrease weight of certain tokens. Examples:
+- _in an autumn park_
+- _in front of a brick wall_
+- _on an ocean beach with (palm trees)++_
+- _in a shopping mall_
+- _in a modern office_
+
+### Additional notes
+
+We use the "same-crop" approach for clothing and avatar images: images will be cropped roughly the same way (using pose 
+estimation), so we don't have to add too much new information (e.g. assume lower body clothing). So, if you use only a 
+photo of an upper body clothing the result will also be cropped the same way regardless of the avatar image (and the 
+other way around):
+
+| Clothing Image                                       | Avatar Image                                        | Result Image                                           |
+|------------------------------------------------------|-----------------------------------------------------|--------------------------------------------------------|
+| <img src="images/clothing_image_02.jpg" width="240"> | <img src="images/avatar_image_02.jpg" width="240">  | <img src="images/same_crop_result_01.jpg" width="240"> |
+| <img src="images/clothing_image_03.jpg" width="240"> | <img src="images/avatar_image_03.jpg" width="240">  | <img src="images/same_crop_result_02.jpg" width="240"> |
+
+## Try-On Output
+
+### Response codes
+
+HTTP status code is used as a high-level response status. In case of a successful API call HTTP code 200 will be 
+returned and response body will contain a resulting JPEG image with the maximum size of 768x1024 pixels. Response
+will also have the "X-Seed" header set that should contain the actual seed used for image generation (for 
+reproducibility). Other status codes (not 200) indicate unsuccessful request, see the table below for additional 
+details:
+
+| Response Code |    Content-Type    |    Headers     | Description                                                                                                                       |                                                    Example                                                    |
+|:-------------:|:------------------:|:--------------:|-----------------------------------------------------------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------------------:|
+|    **200**    |     image/jpeg     | X-Seed: {seed} | Successful API call. Response body contains the resulting image in JPEG format.                                                   |                            <img src="images/same_crop_result_01.jpg" width="160">                             |
+|    **400**    |  application/json  |                | Bad request: at least one of request parameters is invalid. Response body should contain additional error details in JSON format. |                    { "detail": "Invalid upload file type: application/x-zip-compressed" }                     |
+|    **403**    |  application/json  |                | Indicates authentication issue (e.g. invalid API key).                                                                            |                                                                                                               |
+|    **422**    |  application/json  |                | Request validation error. Response body should contain error details in JSON format.                                              |                { "detail": [ { "loc": [ "string", 0], "msg": "string", "type": "string" } ] }                 |
+|    **429**    |                    |                | Too many requests. Might be triggered by the RapidAPI proxy in case of reaching maximum request rate or API call limit.           |                                                                                                               |
+|    **500**    |                    |                | Indicates an internal server error, might not have any details.                                                                   |                                                                                                               |
+
+### NSFW content
+
+We use NSFW content checker to ensure we don't output inappropriate images. If potential NSFW content is detected in the
+generated image, the API will return HTTP status code 400 with a corresponding error message in JSON response.
+
+## Use Cases and Recipes
+
+Our Virtual Try-On API offers a flexible way to specify clothing, avatar and background, which makes it possible to not
+only perform a classic task of virtual try-on, but also generate entirely new images or alter existing images in some
+interesting aspects. Feel free to try and explore!
+
+In all the examples below all unmentioned inputs are assumed to be empty.
+
+### Image-based virtual try-on
+
+The most common use case is to transfer clothing from one photo (e.g. from a product page) to another photo (e.g. 
+user avatar) while maintaining the avatar and the background.
+
+| Clothing Image                                       | Avatar Image                                       | Result Image                                             |
+|------------------------------------------------------|----------------------------------------------------|----------------------------------------------------------|
+| <img src="images/clothing_image_01.jpg" width="240"> | <img src="images/avatar_image_02.jpg" width="240"> | <img src="images/image_based_result_01.jpg" width="240"> |
+
+### Image-based virtual try-on with background
+
+Additionally, it's possible to replace the avatar background with a reference image or a text prompt.
+
+| Clothing Image                                       | Avatar Image                                       | Background Image                                       | Result Image                                                        |
+|------------------------------------------------------|----------------------------------------------------|--------------------------------------------------------|---------------------------------------------------------------------|
+| <img src="images/clothing_image_04.jpg" width="240"> | <img src="images/avatar_image_03.jpg" width="240"> | <img src="images/background_image_01.jpg" width="240"> | <img src="images/image_based_background_result_01.jpg" width="240"> |
+
+And with a text prompt for the background:
+
+| Clothing Image                                       | Avatar Image                                       | Background Prompt            | Result Image                                                        |
+|------------------------------------------------------|----------------------------------------------------|------------------------------|---------------------------------------------------------------------|
+| <img src="images/clothing_image_04.jpg" width="240"> | <img src="images/avatar_image_03.jpg" width="240"> | in front of a snowy mountain | <img src="images/image_based_background_result_02.jpg" width="240"> |
+
+### Avatar from a text prompt
+
+It's possible to replace the person on the clothing image with an avatar, described in a text prompt. Background will be 
+changed as well and will be a random one if not specified:
+
+| Clothing Image                                       | Avatar Prompt                              | Background Prompt  | Result Image                                               |
+|------------------------------------------------------|--------------------------------------------|--------------------|------------------------------------------------------------|
+| <img src="images/clothing_image_02.jpg" width="240"> | a beautiful blond girl with long hair      |                    | <img src="images/avatar_prompt_result_01.jpg" width="240"> |
+| <img src="images/clothing_image_03.jpg" width="240"> | a gentleman with a long beard and mustache | near a fireplace   | <img src="images/avatar_prompt_result_02.jpg" width="240"> |
+
+You may also experiment with avatar prompts for more interesting results:
+
+| Clothing Image                                       | Avatar Prompt       | Background Prompt     | Result Image                                               |
+|------------------------------------------------------|---------------------|-----------------------|------------------------------------------------------------|
+| <img src="images/clothing_image_03.jpg" width="240"> | (iron man mask)+++  | in the Sahara Desert  | <img src="images/avatar_prompt_result_03.jpg" width="240"> |
+
+### Clothing from a text prompt
+
+Similarly, you can specify clothing with a text prompt while providing an avatar image:
+
+| Clothing Prompt                     | Avatar Image                                       | Result Image                                                 |
+|-------------------------------------|----------------------------------------------------|--------------------------------------------------------------|
+| a sheer blue sleeveless mini dress  | <img src="images/avatar_image_02.jpg" width="240"> | <img src="images/clothing_prompt_result_01.jpg" width="240"> |
+| a colorful t-shirt and black shorts | <img src="images/avatar_image_03.jpg" width="240"> | <img src="images/clothing_prompt_result_02.jpg" width="240"> |
+
+### Modifying avatar's body
+
+If you specify clothing and avatar images to be the same while providing an avatar prompt it's possible to change
+avatar's body proportions. Note that it may require using additional term weighting to achieve stronger changes.
+
+| Clothing Image                                       | Avatar Image                                         | Avatar Prompt                 | Result Image                                                     |
+|------------------------------------------------------|------------------------------------------------------|-------------------------------|------------------------------------------------------------------|
+| <img src="images/clothing_image_01.jpg" width="240"> | <img src="images/clothing_image_01.jpg" width="240"> | a (plus size)+ woman          | <img src="images/avatar_modification_result_01.jpg" width="240"> |
+| <img src="images/clothing_image_03.jpg" width="240"> | <img src="images/clothing_image_03.jpg" width="240"> | a (muscular bodybuilder)+++++ | <img src="images/avatar_modification_result_02.jpg" width="240"> |
+
+### Txt2Img
+
+As our diffusion model was fine-tuned to produce people wearing various clothing, it can better follow a clothing prompt 
+and output realistic people and garments:
+
+| Clothing Prompt                                 | Avatar Prompt                  | Background Prompt      | Result Image                                         |
+|-------------------------------------------------|--------------------------------|------------------------|------------------------------------------------------|
+| a paisley pattern purple shirt and beige chinos | a fit man with dark beard      | plain white background | <img src="images/txt2img_result_01.jpg" width="240"> |
+| a white polka dot pattern dress                 | a beautiful petite blond woman | on a yacht             | <img src="images/txt2img_result_02.jpg" width="240"> |
+
+### Other creative possibilities
+
+If you specify the same image for clothing and avatar while providing a background prompt (or background image) you can
+replace the background in a creative way:
+
+| Clothing Image                                     | Avatar Image                                       | Background Prompt       | Result Image                                                |
+|----------------------------------------------------|----------------------------------------------------|-------------------------|-------------------------------------------------------------|
+| <img src="images/avatar_image_02.jpg" width="240"> | <img src="images/avatar_image_02.jpg" width="240"> | on a snowy mountain top | <img src="images/new_background_result_01.jpg" width="240"> |
+
+It's also possible to use a combination of clothing image, clothing prompt, avatar image and a background to add some
+accessories:
+
+| Clothing Image                                       | Clothing Prompt          | Avatar Image                                         | Background Image                                       | Result Image                                                     |
+|------------------------------------------------------|--------------------------|------------------------------------------------------|--------------------------------------------------------|------------------------------------------------------------------|
+| <img src="images/avatar_image_02.jpg" width="240">   | a (light brown purse)+++ | <img src="images/avatar_image_02.jpg" width="240">   | <img src="images/background_image_03.jpg" width="240"> | <img src="images/accessory_result_01.jpg" width="240">           |
+
+## Performance
+
+Typically, one try-on request is processed in 5-10 seconds (depending on type of conditions) excluding network latency.
+In order to reduce network overhead you might want to compress your images before feeding to the API (e.g. using JPEG). 
+Please note that in case of a high demand processing time might increase due to request being queued, though we 
+constantly monitor our GPU cluster capacity and perform scaling as needed.
+
+## Known Issues and Limitations
+
+As any generative model, our models are not perfect (though we constantly work on improvements):
+- Currently, we do not fully support flat lay clothing images. Some might work, but that's not guaranteed.
+- Prompt following might not be perfect, especially in case of long and sophisticated prompts. Prefer simpler and more
+  straightforward prompts whenever possible. Also be pretty verbose (e.g. use the word "plain" if you need something of
+  solid color). Additionally, Compel weighting might be used to increase weight of certain tokens.
+- As usual, generative models struggle with hands, fingers and toes, though we try to mitigate it to a certain extent.
+- Currently, we do not support trying on a single garment, only the full look.
+- Hats and sunglasses are not currently transferred, but we are working on it.
+- Backgrounds might lack some clarity as currently we focus more on clothing.
+- In case of a specified background a hairstyle might change.
+- Body shape of the avatar might change towards smaller sizes.