An attempt at training a model for vtuber logos for flux dev.
Workflow:
Local:
Generate at 1024x512, then 2x latent upscaler.
Tested on Fluxdev-bnb_nf4, works wonders.
