Open davedavis opened 2 years ago
You can increase output resolution by using models that require less VRAM (ViT-B/16 is the smallest, I think) and lowering cuts (for VQGAN+CLIP, trying setting !CUTS=4 in your prompt file). I generally find that using better settings and then upscaling produces much better overall results, though (assuming you're starting from a reasonable output resolution - I try to not drop below 384x384ish).
Cool. Thanks for that.
Would you mind posting the settings (terms used) for the example images? They're amazing.
Unfortunately I made those before I implemented saving the generation settings into the output image metadata. Been pretty busy but whenever I get around to updating this I'll see if I can include some more recent example images with the settings and prompts used.
Congrats, amazing stuff! The results I'm getting are mind blowing!
Unfortunately, like you, I only have an 8GB VRAM GPU. Is there a way to increase the output size to 720 or 1080 without having to upscale? Perhaps by using a smaller chunk size and waiting longer?
Thanks again for releasing this (and for spending the time to create detailed instructions)