ComfyUI Extension: ComfyUI_LucidFlux

Authored by smthemex

Created

Updated

8 stars

ComfyUI_LucidFlux:Caption-Free Universal Image Restoration with a Large-Scale Diffusion Transformer,you can use it in ComfyUI

Custom Nodes (0)

    README

    ComfyUI_LucidFlux

    LucidFlux: Caption-Free Universal Image Restoration with a Large-Scale Diffusion Transformer,you can use it in ComfyUI

    Update

    • 测试环境cu128+torch2.8.0, Vram 4070 12G,Ram 64G ,python3.11,同步官方prompt_embeddings代码,kj dit use links ,官方dit默认要加载clip模型,不建议内存小的人用‘

    1.Installation

    In the ./ComfyUI/custom_nodes directory, run the following:

    git clone https://github.com/smthemex/ComfyUI_LucidFlux
    

    2.requirements

    • 通常不需要
    pip install -r requirements.txt
    

    3.checkpoints

    • any flux dit / 任意flux模型, KJ的 或者官方封装的
    • lucid checkpoints links /lucidflux.pth 和prompt_embeddings.pt(这个是昨天发的 我暂时还没用上)
    • siglip512 links / model.safetensors 只下单体模型
    • DiffBIR links / general_swinir_v1.ckpt
    • turbo lora links #optional 可选,8 步起
    • comfy T5 ,clip-L and flux ae links #comfy T5 ,clip-L is optional / comfy T5和clip-L可选,如图例所示,直接加载emb
    ├── ComfyUI/models/
    |     ├── diffusion_models/any flux dit # 任意flux dit模型 ,就用kj的或者x flux的,名字要带dev 否则跑schnell
    |     ├── vae/ae.safetensors #comfy 
    |     ├── clip/
    |        ├──clip-l.safetensors #comfy optional 可选,如果使用prompt_embeddings.pt
    |        ├──t5xxl_fp8_e4m3fn.safetensors #comfy optional可选,如果使用prompt_embeddings.pt
    |     ├── clip_vision/siglip2-so400m-patch16-512.safetensors  #rename from model.safetensors  最好重命名个,不然都是siglip 的model.safetensors
    |     ├── LucidFlux/
    |        ├──general_swinir_v1.ckpt
    |        ├──lucidflux.pth
    |        ├──prompt_embeddings.pt # 已适配,使用时不要连clip
    

    4 Example

    • use embeddings to save VRAM
    • use clip to encoder prompt

    5 Citation


    DreamClear

    @article{ai2024dreamclear,
        title={DreamClear: High-Capacity Real-World Image Restoration with Privacy-Safe Dataset Curation},
        author={Ai, Yuang and Zhou, Xiaoqiang and Huang, Huaibo and Han, Xiaotian and Chen, Zhengyu and You, Quanzeng and Yang, Hongxia},
        journal={Advances in Neural Information Processing Systems},
        volume={37},
        pages={55443--55469},
        year={2024}
    }