>>84881958https://files.catbox.moe/6ymxvp.pngMy current workflow seems to be too much for my gpu. So I've mostly just been testing stuff, not that much to show for it.
Also, I've noticed that any diffusion process I start has a chance to just slowly use more and more vram, it doesn't matter if comfy, reforge or kohya, they all slowly increase in vram usage until I have to restart the process to reset it. Is that normal, or an AMD issue?
>>84884377Sure, here you are
https://files.catbox.moe/odpp3s.safetensors>>84883808Hm, i'm stuck at around 15-20s/it on my 6800xt, even with Q4. And I'm finding mixed info, with some people reporting nearly 5s/it and others having exactly my speed. Gonna try to figure out if maybe the issue is the gguf implementation, but I doubt it. The fp8 version still needs a bit to download though.
>>84891092Very nice!
>>84905454Nice bread!
>>84933464Recursion! Very cute style
>>84933864>4th tail 5.0Seems like it gets anatomy a bit more correct from my short tests.
>>84939391>picrelHoly, that looks good. Is that just flux with Lora, or a second pass with an XL/Pony model?
