/vt/ - Virtual Youtubers

>>56738401
I'm not an inpainting expert or anything but here's what I do which I guess covers the basics -
1) Put your image into inpaint (not inpaint sketch)
2) Figure out a prompt. Personally I like to use the negatives from the original image, as well as the parts of the original prompt that are relevant to the inpainted section (so if you're inpainting a face remove your prompts about their shoes or whatever). You might have to add things to negatives if it isn't inpainting well
3) Mask the area you want to inpaint. Best to keep it smaller if possible, so if you want to inpaint a hand and a face on opposite sides of the image, do it one at a time
4) Change inpaint area to "only masked" and turn the padding up a lot, probably just max it
5) Width and height should be fine at 512x512. If your area is a lot larger in one direction than the other you can try to match it some but you shouldn't need to change the resolution to anything super high.
6) You have a lot of freedom with the denoising strength slider. The higher the value is, the more the AI is allowed to change what was originally there. There's not really a "correct" value imo because it depends on how close the input is to what you want so you just have to experiment.

And that's it really. You can try different samplers, different steps (I find higher step counts are a bit more reasonable here since you gen at low resolutions which makes it a bit quicker), CFG scale, all that stuff. Personally I tend to gen them in batches of 4 and just kinda compare them all to see if the spot got any better or not.
If you want to make big changes to the image or want something very specific, you might have to doodle on it a bit in paint first as guidance.