/vt/ - Virtual Youtubers

>go to load a different model
>webui has been keeping all models in system RAM
>even after switching models
>and not letting go
fuck it, I'm gonna go steal some patches for the Colab, this shit is so annoying

>>38847479
nothing, was just wondering if it'd be worth trying later or not when I'm physically at a display with accurate colors. I enhance'd a few times and I like different bits of each one. prob gonna try playing with layers/blending/masking to make a combination out of them, if they actually line up.
>>38847523
I should've added on lol, I quite like how it makes things turn out because I can turn scale down to 8 or 9 and still get an image that looks as sharp as one done on scale 14+, but with an even lower scale's style of composition. wish I understood how "unconditional conditions" and the positive proompt's tug of war actually worked in relation with scale. I didn't remove anything from it though, been using it as is on top of the default excludes which I left in first
>>38847703
length definitely does, but it seems to work the opposite way from making a bloated ass 300 token positive prompt, where you would actually want to turn scale up. making the negative prompt bloated seems to act like turning scale up.
>>38847500
>image of reine tripping balls
I understood...some of that. but the t12 is supposed to be the new and improved? where'd the cheese go? the cheese was sovl.
>batching
so it does 1 step per image and cycles through them 5 at a time? like when it batches output?
>LR
fuck Adobe Lightroom
>gradient accumulation
nahone
is there a resource I can go read?