-
Beta Was this translation helpful? Give feedback.
Replies: 3 comments 4 replies
-
I don't use the Comfy WebUI so I don't know the answer to your 1st question. As for upscaling manually, I didn't see blur or artifacts, I upscaled with one of the things you mentioned under 3. In the interface I have 4 options for upscaling. I never saw the interface without the optional components so I don't know whether there is a "low quality" upscaler available when you don't install that HAT and Real HAT GAN. As for your 3rd question, when you let the plugin install everything all content can be found under |
Beta Was this translation helpful? Give feedback.
-
All highres fix is: Base image gets generated at a smaller resolution (512x512 by default for SD 1.5) > base image gets upscaled using the selected upscaler to the target resolution (let's say 2x so 1024x1024) > upscaled image gets run through img2img in the background to refine it > result You can replicate this in Krita Diffusion using the "Upscale" mode and enabling "Refine upscaled image". The strength slider is effectively denoising strength. Or manually upscale an image using Krita's built in Image > Scale Image to New Size, then in the default "Generate" mode lower the strength slider below 100% and the button should change to "Refine" and then use that. |
Beta Was this translation helpful? Give feedback.
-
Hires fix is done automatically. Just generate at any resolution, it will use two-pass generation if the resolution is outside of the checkpoint's recommended range. The "Upscale" tab is useful for very high resolutions (it processes the image in tiles) or if you want to select a custom upscale model. |
Beta Was this translation helpful? Give feedback.
Hires fix is done automatically. Just generate at any resolution, it will use two-pass generation if the resolution is outside of the checkpoint's recommended range.
The "Upscale" tab is useful for very high resolutions (it processes the image in tiles) or if you want to select a custom upscale model.
You can download HAT models from here: https://huggingface.co/Acly/hat/tree/main