Flux Inpaint and Outpaint Workflow版本SDXL Outpaint (ID: 940708)

Notice

• This workflow is not state of the art anymore, please refer to the Flux.1 Fill and the official comfyui workflows for your inpainting and outpainting needs. Details below:

• Black forest labs has since released official flux 1 tools: https://blackforestlabs.ai/flux-1-tools/

• The tools include the Flux.1 Fill [dev] model which is better than the alimama controlnet used in this workflow.

• There is native comfyui support for the flux 1 tools: https://blog.comfy.org/day-1-support-for-flux-tools-in-comfyui/ which you should use for your inpainting and outpainting needs.

• You can grab the official comfyui inpaint and outpaint workflows from: https://comfyanonymous.github.io/ComfyUI_examples/flux/ by downloading the images (workflow embedded within the images) and opening them in comfyui.

• If you have low VRAM, you can download the fp8 version of the Flux.1 Fill [dev] model from here: https://civitai.com/models/969431/flux-fill-fp8

• This workflow will not be updated anymore because there is nothing I have to add to the official comfyui inpaint and outpaint workflows.

Introduction

The latest version of this workflow uses

• alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta

• alimama-creative/FLUX.1-Turbo-Alpha

to achieve 8 steps inpainting and outpainting within the same workflow.

Models

1. FLUX.1-Turbo-Alpha.safetensors (models/loras/flux) https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha

2. FLUX.1-dev-Controlnet-Inpainting-Beta-fp8.safetensors (models/controlnet) https://huggingface.co/alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta

   Download diffusion_pytorch_model.safetensors, rename the file and use Kijai's script (https://huggingface.co/Kijai/flux-fp8/discussions/7#66ae0455a20def3de3c6d476) to convert to FP8 to be able to fit into 16GB VRAM.

3. flux1-dev-fp8-e4m3fb.safetensors (models/diffusion_models/flux): https://huggingface.co/Kijai/flux-fp8/tree/main

4. t5xxl_fp8_e4m3fn_scaled.safetensors (models/clip): https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main

5. ViT-L-14-BEST-smooth-GmP-TE-only-HF-format.safetensors (models/clip): https://huggingface.co/zer0int/CLIP-GmP-ViT-L-14/tree/main

Custom Nodes

1. segment anything (if you have > 16GB VRAM and want to use automatic segmentation)

2. Various ComfyUI Nodes by Type

3. KJNodes for ComfyU

描述:

Updated this workflow to use xinsir's SDXL Union Controlnet

训练词语: