>>39632198https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/5521TL;DR it lets you manually get something close to "hi-res fix" but on img2img upscales instead of being limited to using it initially with txt2img, which opens the door for things like using "hi-res fix lite" after you inpaint, or after switching models, or changing the prompt a bit, etc.
someone PR'd a commit that gives you a new option on the img2img tab - which normally does any upscaling in "image" space. Instead it tokenizes the image again and does the upscaling in black magic AI CLIP "latent space".https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/5840 you can also use it in x/y plots now too
https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/5976 - I've been using this on my Colab btw, I thought it was always there but I guess I was mistaken. it's quite nice, but the previews are "your VAE is off"
>>39632306>lam, mikapikazonani sore?