Upscaling image with SD3.5 Large up to a size of more than 1 megapixel.
We'll use imgtoimg, this controlnet Blur: https://civarchive.com/models/984199?modelVersionId=1103617
And this extension (Tiled Diffusion)
With the help of tile splitting, you can overcome the limitations of SD 3.5 in 1 Megapixel.
There is recommendation: tiling the image at a tile size between 128 and 512. But i got better result with preset parameters of node, and other parameters took from post of aeroumbria на одном ресурсе.
You can try different parameters and achieve much better results, it takes me a very long time to do this because of the old CPU and 8GB of VRAM. You can use any model based on SD 3.5 Large. I also tried Absynth - SD3.5L and SD3.5-SLG-Weighted-Merge GGUF version.
You can also inpaint without upscale, it turns out beautifully.
The images that are submitted for imgtoimg and for the controlnet input must be the same size. To use controlnet, you need the latest version of ComfyUI. If you get an error, update ComfyUI.
Description
FAQ
Comments (3)
I don't understand how to use this. There are two groups with a load image node. Do I load the same image into both? Only use one or the other? This is new to me sorry.
So I'm appreciating this, but this controlnet was supposed to work up to 8k resolution. You're only upscaling to 1536 res which is barely beyond SD 3.5's normal render point. What about 3k and beyond? What settings would we use for that? Thanks.
Lovely! Thanks a bunch for sharing. I'm new to AI gen stuff, and I think this is exactly what I was looking for. Cheers!
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.