/vt/ - Virtual Youtubers

>>60087106
>>60087141
It's somewhat fixable. I can train on each image twice: one with old tags, one with new descriptions. Can ask it to generate shorter descriptions or use something like token_lenght 300 with clip skip 1
I'll try retagging few thousand images like this and making something. It will be bad, but trying to prompt "possibly a cat" yukimins and "possibly an elf" Flare should be fun.