Initial commit

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* 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

@@ -1,13 +1,7 @@
----
-title: Try On Diffusion
-emoji: 🚀
-colorFrom: blue
-colorTo: pink
-sdk: gradio
-sdk_version: 5.3.0
-app_file: app.py
-pinned: false
-short_description: Diffusion-based multi-modal virtual try-on demo
----
+# Virtual Try-On Diffusion [VTON-D] by Texel.Moda
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+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. Feel free to [contact us](https://texelmoda.com/contact/) if you are interested in commercial API usage or 
+partnership.

app.py

@@ -0,0 +1,276 @@
+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", "")
+
+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_RGB2BGR), 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()) 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. 
+            Feel free to <a href="https://texelmoda.com/contact/" target="_blank">contact us</a> if you are interested 
+            in commercial API usage or partnership.
+        </p>
+        <br/>
+    """
+    )
+
+    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", elem_id="button")
+
+            result_image = gr.Image(label="Result", show_share_button=False)
+            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)

requirements.txt

@@ -0,0 +1,4 @@
+requests==2.32.3
+requests_toolbelt==1.0.0
+numpy==1.26.4
+opencv-python-headless==4.10.0.84

try_on_diffusion_client.py

@@ -0,0 +1,119 @@
+import cv2
+import numpy as np
+import requests
+from requests_toolbelt.multipart.encoder import MultipartEncoder
+import logging
+import json
+from io import BytesIO
+from dataclasses import dataclass
+
+
+@dataclass
+class TryOnDiffusionAPIResponse:
+    status_code: int
+    image: np.ndarray = None
+    response_data: bytes = None
+    error_details: str = None
+    seed: int = None
+
+
+class TryOnDiffusionClient:
+    def __init__(self, base_url: str = "http://localhost:8000/", api_key: str = ""):
+        self._logger = logging.getLogger("try_on_diffusion_client")
+        self._base_url = base_url
+        self._api_key = api_key
+
+        if self._base_url[-1] == "/":
+            self._base_url = self._base_url[:-1]
+
+    @staticmethod
+    def _image_to_upload_file(image: np.ndarray) -> tuple:
+        _, jpeg_data = cv2.imencode(".jpg", image, [int(cv2.IMWRITE_JPEG_QUALITY), 99])
+        jpeg_data = jpeg_data.tobytes()
+
+        fp = BytesIO(jpeg_data)
+
+        return "image.jpg", fp, "image/jpeg"
+
+    def try_on_file(
+        self,
+        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,
+        negative_prompt: str = None,
+        num_images: int = 1,
+        seed: int = -1,
+        raw_response: bool = False,
+    ) -> TryOnDiffusionAPIResponse:
+        url = self._base_url + "/try-on-file"
+
+        request_data = {"num_images": str(num_images), "seed": str(seed)}
+
+        if clothing_image is not None:
+            request_data["clothing_image"] = self._image_to_upload_file(clothing_image)
+
+        if clothing_prompt is not None:
+            request_data["clothing_prompt"] = clothing_prompt
+
+        if avatar_image is not None:
+            request_data["avatar_image"] = self._image_to_upload_file(avatar_image)
+
+        if avatar_prompt is not None:
+            request_data["avatar_prompt"] = avatar_prompt
+
+        if avatar_sex is not None:
+            request_data["avatar_sex"] = avatar_sex
+
+        if background_image is not None:
+            request_data["background_image"] = self._image_to_upload_file(background_image)
+
+        if background_prompt is not None:
+            request_data["background_prompt"] = background_prompt
+
+        if negative_prompt is not None:
+            request_data["negative_prompt"] = negative_prompt
+
+        multipart_data = MultipartEncoder(fields=request_data)
+
+        try:
+            response = requests.post(
+                url,
+                data=multipart_data,
+                headers={"Content-Type": multipart_data.content_type, "X-API-Key": self._api_key},
+            )
+        except Exception as e:
+            self._logger.error(e, exc_info=True)
+            return TryOnDiffusionAPIResponse(status_code=0)
+
+        if response.status_code != 200:
+            self._logger.warning(f"Request failed, status code: {response.status_code}, response: {response.content}")
+
+        result = TryOnDiffusionAPIResponse(status_code=response.status_code)
+
+        if not raw_response and response.status_code == 200:
+            try:
+                result.image = cv2.imdecode(np.frombuffer(response.content, np.uint8), cv2.IMREAD_COLOR)
+            except:
+                result.image = None
+        else:
+            result.response_data = response.content
+
+        if result.status_code == 200:
+            if "X-Seed" in response.headers:
+                result.seed = int(response.headers["X-Seed"])
+        else:
+            try:
+                response_json = (
+                    json.loads(result.response_data.decode("utf-8")) if result.response_data is not None else None
+                )
+
+                if response_json is not None and "detail" in response_json:
+                    result.error_details = response_json["detail"]
+            except:
+                result.error_details = None
+
+        return result