ten-c++ 🚀❤️‍🔥 ✂️✨ (_/acc)さんのイラストまとめ

ロケットアイドルくりやーちゃん！！！ #VTuber #女の子 #宇推くりあ https://t.co/FNkAAvFESz

Stopping training for the 128x128 model here, it's gotten good enough to prove the architecture works, now to work on increasing resolution and reducing training time.

Attempting to find close representations to a target image doesn't seem to go very well, some major features are reflected in the top rated results of each iteration, so maybe I should explore using a more creative search approach

this culminates in the following 128x128 output:

And in animation form:

Something I'm working on ;)

Dramatic improvement in generated quality via feature matching

Testing out a 128x128 to 512x512 upscaler

4-5 hours of transfer learning from my previous best GAN on a newer dataset