Search Results
7/3/2025, 1:50:22 AM
>>101869046
I think most everybody got to it before me kek but yeah same as everyone said upscale then inpaint. The limitation with XL models is that it has pretty bad vae so it can't really do small fine details so the best way to get those is to upscale (essentially making those fine details bigger) and inpaint at that higher res.
>>101869803
I think you got it but just to add a few details to those settings (denoising strength and inpaint only masked should be straight forward)
>mask blur
Basically it adds a slight blur around the inpaint area to help it mask with it's surrounding better, was more important in the 1.5 era, XL is better at meshing them together so you can keep it at 0.
>height/width
Takes your masked area and upscales it to the set height width before inpainting
>Masked padding
This imo is the most important thing to understand in inpainting. When you mask and send to the model to inpaint you are basically ripping that masked portion out and showing it to the model to inpaint on, it can't see any more of the image only the masked portion. So what padding does is it takes your masked area and expands the area it's gonna rip out (bigger the number, bigger the portion it will cut out of the image) but only the masked portion will change the rest of the padding is sent as context to help the AI understand what it is inpainting on.
So inpainting is a balancing act between the denoise, padding area, high denoise with low padding is when you get people inside faces situation, because AI doesn't have enough context on what it is inpainting on.
I think most everybody got to it before me kek but yeah same as everyone said upscale then inpaint. The limitation with XL models is that it has pretty bad vae so it can't really do small fine details so the best way to get those is to upscale (essentially making those fine details bigger) and inpaint at that higher res.
>>101869803
I think you got it but just to add a few details to those settings (denoising strength and inpaint only masked should be straight forward)
>mask blur
Basically it adds a slight blur around the inpaint area to help it mask with it's surrounding better, was more important in the 1.5 era, XL is better at meshing them together so you can keep it at 0.
>height/width
Takes your masked area and upscales it to the set height width before inpainting
>Masked padding
This imo is the most important thing to understand in inpainting. When you mask and send to the model to inpaint you are basically ripping that masked portion out and showing it to the model to inpaint on, it can't see any more of the image only the masked portion. So what padding does is it takes your masked area and expands the area it's gonna rip out (bigger the number, bigger the portion it will cut out of the image) but only the masked portion will change the rest of the padding is sent as context to help the AI understand what it is inpainting on.
So inpainting is a balancing act between the denoise, padding area, high denoise with low padding is when you get people inside faces situation, because AI doesn't have enough context on what it is inpainting on.
Page 1