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.