Model description
Trained and hand tuned for Monochrome manga panels and art.
a full FLUX-DEV checkpoint rather than just loras because best results were observed using this method.
naturally has consistant characters across panels without requesting it.
14 Plus Loras all 32 rank merged into FLUX-dev base all at variying weights all hand dialed into to acheived dersired aesthetic outcomes.
over 36,000 training steps total.
I also added the indepdant loras to add and mess around with yourself but do not expect the checkpoints quality out of the box with the indepdant loras since it took me a lot of tinkering to get that quality with a mix of the loras.
shoutout to FAL for sponsering me with compute credits and making this possible!
Trigger words
Technically it was not trained with a trigger word but using
monochrome manga
is highly suggested
Use it with the :firecracker: diffusers library
from diffusers import AutoPipelineForText2Image
import torch
pipeline = AutoPipelineForText2Image.from_pretrained("dataautogpt3/FLUX-MonochromeManga", torch_dtype=torch.bfloat16).to('cuda')
image = pipeline('Monochrome Manga of a girl').images[0]
License
Please adhere to the licensing terms as described here.