what's the difference with V1?
#8
by
flankechen
- opened
https://huggingface.co/XLabs-AI/flux-ip-adapter
any more tech detail or report?
So, 500k steps vs 75k, and 13x larger dataset, 16 visual tokens instead of 4 in v1
So, 500k steps vs 75k, and 13x larger dataset, 16 visual tokens instead of 4 in v1
thanks, is the clip image projector still linear+layernorm as the original ipadapter paper base model?
would you try to train with plus like, resampler model?
no, only default version