/vt/ - Virtual Youtubers

The wide anon is on the money. Here's a more mundane example:

I generated this Irys, but her face looks like garbage. And her hands are fucked up, too. Very simple fix.

I mask her face. I set inpaint area to 'only masked', max padding, and resolution to at least 704x. This tells SD to regen the masked area at 704x instead of 200x or whatever the size of the masked area originally was, using as much context as it can get from the area around it. This makes it super easy to get detailed faces, hands, clothes, etc, while keeping the rest of the pic intact. I set denoising to .6 so it actually has room to generate something new.

I inpaint her face, I like what I got, send it back to inpaint. One hand after the other, boom, we got a finished pic. I'll give her a cameltoe too for funsies.

It's so easy that I don't know why more people don't do it. It would take a hell of a lot less time than generating until you found something nice

https://files.catbox.moe/m1jmrf.png