v2:
retrained with new gemma 4 captions- hopefully to have better prompt adherence as gemma 4 seems to be a step up from qwen 3 vl i used before.
v1.
Works
Strength between .5 and 2
Description
FAQ
Comments (7)
thanks for the model!
any chance we could get one for the belt only?
prompt for it? most of the dataset is belt only
@Joschek have tried, but keep getting chains on the side
@rsztjhm I'll test and maybe theres a chance I do a run with a reduced dataset, but don't hold your breath
@rsztj hm you are correct, it does have this bias, ill put it on a list of Lora I wanna retrain ...
@rsztj i recaptioned the dataset and retrained. i think its gotten better. let me know if it does for you if you happen to test it out
@Joschek the new version doesn't have that issue anymore, thanks!