Shawn Presserさんのイラストまとめ

Anyone know which anime this is? Oh, right, it’s *computer generated*. @AydaoGMan is a true stylegan master.

Never seen stylegan make letterforms before.

After many, many months, @AydaoGMan has finally cracked “stylegan on complex datasets”. Look how good these are! Trained on 2M danbooru photos, I think.

I can’t overstate how much work this took. There are so many variants to try that I basically gave up.

After all, StyleGAN's architecture is quite similar to BigGAN. So it must be due to one of the "bells and whistles" that StyleGAN comes with.

I'll just give the answer: Style mixing on (left, step 26k), style mixing off (right, 24k). Note how much more cohesive the bodies are.

This pretty much sums up my last few months in AI. At this point I’ve become desensitized for life in certain aspects.

Also this is my fursona.

BigGAN 256x256 on danbooru2019figures (https://t.co/y2eii80IfN)

step 113k, halfway through training.

self-attention @ 128x128; 96ch for both G and D; EMA samples.

AI generated (BigGAN)