ComfyUI Extension: ComfyUI-DiffusersLoader

Authored by Scorpinaus

Created

Updated

14 stars

This node pack allows loading of SD checkpoints that uses diffusers format in comfyUI.

Custom Nodes (0)

    README

    Diffusers Format Checkpoint Loaders for ComfyUI

    This project aims to create loaders for diffusers format checkpoint models, making it easier for ComfyUI users to use diffusers format checkpoints instead of the standard checkpoint formats.

    This project was created to understand how the DiffusersLoader available in comfyUI works and enhance the functionality by making usable loaders.

    Table of Contents

    Introduction

    This project provides tools and loaders for integrating diffusers format checkpoint models with ComfyUI. It helps users who already have diffusers format checkpoints to seamlessly use them without needing to convert them to the standard checkpoint format.

    Installation

    Follow these steps to install and set up the project:

    1. Download the node via ComfyUI-Manager by searching: DiffusersLoader: image

    2. Add nodes to your workflow by:

    • Add Nodes -> DiffusersLoader -> Combined -> CombinedDiffusersLoader
    • image
    1. Connect the DiffusersLoader to your workflow

    Features

    The CombinedDiffusersLoader supports loading of diffusers checkpoints for:

    • SD 1.5
    • SD 2.1
    • SDXL
    • SD3
    • AuraFlow
    • Flux

    Supports loading via extra_model_paths.yaml - Using ComfyUI

    • If there are 2 model directories of the same name in different places, the model_name(1) refers to the directory in your 1st base path while model_name(2) refers to the directory in your 2nd base path.

    Workflow Example

    The combined loader work in the same manner as existing checkpoints loader as seen in this workflow:

    image

    Select the name of the diffusers checkpoint folder from the dropdown list and connect the nodes.

    Limitations & Future Improvements

    • Add support for other compatible diffusers format checkpoints
      • Future model_types:
        • HunyuanDiT
        • PixArt
        • Kolors Note: This will need to raise a PR on main comfyUI repo
    • For Flux, use of default weight_dtype would OOM (Use fp8-based d-types)

    Contributing

    Contributions are always welcome! Please follow the contributing guidelines to submit issues or pull requests.

    License

    This project is licensed under the MIT License.

    Contact the Author

    Hafiz Saffie - [email protected]

    Project Link: https://github.com/Scorpinaus/ComfyUI-DiffusersLoader

    Update History

    26/8 - Improved on model detection to include extra_model_paths.yaml. 25/8 - Implemented multi-file loading support where the combined_safetensor.safetensor file method is used as a back-up 21/8 - Compatibility Update With ComfyUI, Addition of Flux Support 30/7 - Added repo to comfyUI registry 28/7 - Added AuraFlow support 20/6 - Added SD3 support 2/6 - Initial commit