UPDATE
Successor model can be found here and will be much better for foot focus generations and inpainting help!
https://civarchive.com/models/189085/mizars-foot-focus-helper
___________________________________________________________________________________________
I've experimented use with weights 0.45 - 0.70 when inpainting and 0.40 - 0.45 using Txt2Img generation -- it works consistently well in both cases. I use a CFG scale of 7, I don't think there's really a need to increase it beyond that. Low denoise strengths for Img2Img inpainting (0.4 - 0.6) help maintain nail polish color if your image has any, as it tends to skew polish at high percentages.
I trained this LORA on closeups with the intention of training a separate one for tandem use that'll handle general foot focus poses with a body -- This is mainly for corrections and texture fixes. It works with 'soles' and downward 'pointed toes', granted that you specify either in the prompt (DO NOT USE BOTH). You should specify 'Left Foot' or 'Right Foot' based on which is being inpainted for the best results. Left/Right are from model POV, not viewer POV.
Recommended Keywords:
barefoot, feet, toes, close-up, foot focus, facing viewer, right foot, toenail polish, toenails
Specify 'soles' or 'pointed toes' based on what is being inpainted.
Specify 'Left foot' or 'Right foot' if inpainting individually
Nail polish color can vary, this was trained on Black, white, purple, and aqua.
Foot focus LORA that is trained on full-body poses soon hopefully, let me know well how this one works. If you like working with it please consider supporting on Patreon or tipping buzz here so I can train more. My GPU is weak and I trained this one via CivitAi's training option.
Description
FAQ
Comments (3)
To be honest for 100 renders I have like 1-2 good feet renders. Looks like working bad with most popular anime models.
Care to share the actual use cases? Inpainting, Txt2Img generation? This is mainly for low denoise strength inpainting for texture and digit fixes. Based on your prior works, if you were trying to inpaint pointed toes, it's less consistent on that front due to being trained more on soles and works better in that respect. There will be another version with more training but I need more sample images to go off of (and buzz)
@MizarAiArt I used inpaint area "only masked" like I see that was mistake, with "whole picture" I have better renders. looks good! I combo ur lora and negfeet embe and its looking great!