>>48441186>>48441261I use DPM++ 2m Karras with txt2img. Then I do a multi-diffusion, Noise Inverted Upscale in img2img. Noise Inversion seems to only want to use Euler. Here's the same image using Multidiffusion but without Noise Inversion, which hooked into the DPM++ 2M v2 Sampler that I have.
>>48441512I did, I can upload it again if you would like.
PixelDrain is being annoyingly slow in uploading/calculating parity data. I'll reply again when I have a link.
>>48441431Bayesian Merger automates a lot of the process by essentially trying to optimize to the highest score. If you manually score, you can force the optimizer to converge to a specific thing that you are looking for in the merge.
>>48441769>Imma keep it real with you, I didn't understand half of what you just said, but it sounds cool.That's fine. All you really need to know is that it makes block merging less tedious, and most likely gets better results over both brute-forcing and manually testing block weight merges.
