ComfyUI Extension: ComfyUI_sm4ll-Wrapper

Authored by risunobushi

Created

Updated

1 stars

A ComfyUI custom node that integrates with the sm4ll-VTON API for virtual try-on functionality.

Custom Nodes (0)

    README

    ComfyUI_sm4ll-Wrapper

    A wrapper node for sm4ll-VTON models: https://huggingface.co/spaces/sm4ll-VTON/sm4ll-VTON-Demo

    Installation

    1. Clone or download this repository to your ComfyUI custom nodes directory:

      cd ComfyUI/custom_nodes/
git clone https://github.com/your-repo/ComfyUI_sm4ll-Wrapper.git

    2. Install the required dependencies:

      cd ComfyUI_sm4ll-Wrapper
pip install -r requirements.txt

    3. Restart ComfyUI

    Usage

    1. Add the Node: In ComfyUI, look for "sm4ll Wrapper Sampler" in the sm4ll/VTON category

    2. Configure Inputs:

      • base_person_image: The person/model image (required)
      • product_image: The garment/product to try on (required)
      • model_choice: Select from "eyewear", "footwear", or "full-body" (required)
      • (optional) mask_image: if you don't trust the automasking feature, provide a mask drawn where the product should appear on the person

    3. Connect Output: The node outputs a processed IMAGE that can be connected to other nodes

    How It Works

    1. Image Processing: Input images are automatically resized to 1.62 megapixels using Lanczos interpolation for optimal results
    2. Upload: Images are uploaded directly to the Gradio space's internal file system
    3. API Call: The VTON API is called using the uploaded file paths with proper session management
    4. Result Retrieval: The resulting image is downloaded and converted back to ComfyUI tensor format

    Technical Details

    • API Integration: Uses the official sm4ll-VTON Gradio API with proper SSE streaming
    • Session Management: Maintains session state for reliable file uploads and downloads
    • Error Recovery: Includes comprehensive error handling and fallback mechanisms
    • Format Support: Handles various Gradio response formats and URL patterns

    Troubleshooting

    • Red Output Image: Indicates an error occurred during processing. Check the console for detailed error messages.
    • Timeout Issues: The API has a 5-minute timeout for processing. Large images or high server load may cause timeouts.
    • Upload Failures: Check your internet connection and verify the Gradio space URL is accessible.

    Requirements

    • ComfyUI
    • torch
    • torchvision
    • numpy
    • Pillow
    • requests

    License

    This project is licensed under the MIT License - see the LICENSE file for details.

    Credits