Flux Fill Inpainting With SD 1.5 Refiner版本v0.6 (ID: 1161897)

This is a somewhat simple workflow that takes full advantage of the flux teams new fill model. This is the best implementation I have seen of it so far. The biggest issue with flux fill is it overly smooths everything so I used a SDXL refiner to try and add that realism and detail back and it came out incredibly well. It also has an automasking feature using grounding dino sam that works pretty well and saves me inpainting time. It also has native Lora support so you can get some really specific usecases out of it. So far I haven't seen any custom checkpoints of flux fill so I'm using stock but I do hope that changes soon.

This is my first project and I wanted to design it to be user-friendly and simple and I hope I accomplished that. There is a super long note that trys to explain how to use it, but if you have used comfyui before then I think you should be able to understand it pretty quick. It uses a load of custom nodes and a lot of models that I'll list here :
Flux Model:

• flux1_fill_dev

1. VAE Model:

   • flux_vae (ae.safetensors)

2. Load Clips Node (Flux):

   • clip_l

   • t5xxl_fp8_e4m3fn

3. Grounding Dino Model:

   • GroundingDINO_SwinT_OGC (694MB)

4. Sam Model:

   • sam_vit_h (2.56GB)

5. SD 1.5 Inpainting Model:

   • Any SD 1.5 inpainting model works fine.

   • Recommended: Realistic Vision Inpaint

Please leave any feedback or questions below I will try to answer as best I can. The workflow is still in beta but this is the 6th iteration of the project so far and I think its in a shippable state. Thanks!

描述:

• First version of release

• Reworked The UI to user friendly

• Added Flux Support

• Added SDXL Refiner

• Added AutoMasking

• Fixed VAE bug that made output images look smushed

• Added comparisons

• Added Native Lora Support

• Adding masking options

训练词语: