Trained on 369 screen captures from Golden Boy at native res 768x576.
H3 is wonderful, it knows anime quite well, I will potentially try training it on video clips from the show but it does a pretty good job with just images.
It trains mega fast on images at this res, around 1.1s/it and this lora trained in around 3 hours. I'll do some extra training tomorrow but I am happy enough with the 12k checkpoint.
Trigger:
retro 90s 2d anime style, golden boy style
Not tested with any speed up loras like lightx. All examples are t2v 1mp 20 steps, euler beta57.
Training:
Trained using musubi tuner fork, lr 5e-5, --h3_spatial_density_jitter 0.2 --h3_guidance_distillation_scale 3.5 --h3_base_preservation_loss_weight 0.02 \
Rank 16
Description
FAQ
Comments (8)
Are you serious?? That hair animation on the first example video! It looks like it's actually drawn and animated! These video gen models tend to want to turn everything into a 3D model, or do something that looks like rigged animation. Example here for Wan, regarding hair being treated as a 3D model.
I am floored that you trained this in 3 hours, and only on images. SO happy that Minimax has such a great base understanding of this stuff. It opens up so much. The resulting loras are smaller than previous video models as well.
Yeah this is truly the first video model properly trained on anime. Very excited for the future. H3 is a bigger smarter model, lower rank will do fine so smaller file sizes
@tazmannner379 I want to train an X-Men 97 lora, though I'm afraid that images won't be enough. But yeah, for 3 hours, it's so worth trying. Thanks for sharing training info.
Golden Boy was a great choice for this look. Well done.
Boys! Its here! 😎
😍 👍🏻
Excellent work
bro! You are my hero.