Looks fantastic! Unfortunately I don't have the hardware for ai image generators but I do use cloud computing resources. Are there any google colab notebooks available for this model?

(March 12, 2024, 5:21 pm)Bred_anonymously Looks fantastic! Unfortunately I don't have the hardware for ai image generators but I do use cloud computing resources. Are there any google colab notebooks available for this model?

Not sure if Google Colab would have the compute, least not the free tier. I got it running on runpod before I splashed out on more RAM (needs over 32GB like).

Used this fella's installer for runpod since i'd never touched before.... although it's behind a patreon:

hxxps://huggingface.co/blog/MonsterMMORPG/supir-sota-image-upscale-better-than-magnific-ai

(March 12, 2024, 3:46 pm)Mothman125 I tried to look up what the control scale does exactly and didn't easily find an answer.

Basically the same as a controlnet strength, over 1.00 and a lot more of the original image is included (or whatever the adapter is pulling), then starts getting burnt.

Pretty sure using zero control would be essentially just a pure SDXL generation, again much the way of a controlnet.

I've never messed with controlnet settings. Only ever used controlnet to inpaint or make SDXL inpaint worth a shit, which I haven't figured out yet.

(March 12, 2024, 8:23 pm)mimmiberg

(March 12, 2024, 5:21 pm)Bred_anonymously Looks fantastic!  Unfortunately I don't have the hardware for ai image generators but I do use cloud computing resources.  Are there any google colab notebooks available for this model?

Not sure if Google Colab would have the compute, least not the free tier. I got it running on runpod before I splashed out on more RAM (needs over 32GB like).

Used this fella's installer for runpod since i'd never touched before.... although it's behind a patreon:

hxxps://huggingface.co/blog/MonsterMMORPG/supir-sota-image-upscale-better-than-magnific-ai

I have the paid tier of colab so I've tried higher end models like SDXL in it before.  I suppose it should be possible to run this if I could get a notebook set up

I threw a 4090 at it and it still takes a couple minutes.

Apparently should have support for the Lightning SDXL models now (specifically mentions Juggernaut XL Lightning in latest commits) which could provide a decent speed-up, although they require a much smaller CFG scale, so not sure how that'll look over 4~8 steps.