//=time() ?>
マヤちゃん誕生日おめでとう!
最推しウマ娘です
https://t.co/1FOSSHjaSX
#ウマ娘
#マヤノトップガン
#マヤノトップガン生誕祭2023
#AIart #AIイラスト
The thing is that you can get basically characters that appear in anime rendered quite correctly.
After training for 33 hours I still cannot get most of the outfits correct but I guess you can prompt almost every character without problem
Intermediate checkpoints are here:
https://t.co/klMqNH660c
#ウマ娘
#stablediffusion
#AIart #AIイラスト
Style can be tweaked as well. All right, I hope the outfits and the remaining characters can be trained into the model with enough patience.
For some reason some characters are not yet trained into the model. I hope the shuffling is effective because my training hangs at epoch 1/3 so I resume with another random seed.
Will also need more time to improve multi-character generation and this is where we are.
Intermediate results after 15 hours of training.
Clothes need more time to bake in but due to the limitation of clip some of them will probably fail.
A better way would be to do embedding + model training.