Works "good" with edit, The idea is using with inpaint to increase detail.

This is far from perfect:

• Only works from the back
• May try to replace any hole with an or a
• May create some weird nightmare images

I'm releasing because may be useful for someone...

The lora trained on Musubi-Tuner is better for text-to-image.

The lora trained on AI-Toolkit is better for image edit.