minimax h3 plz
minimax h3 is better than all video gen models open source before
minimax h3 is better than all video gen models open source before
Yes, definitely, it natively does head swapping but it's not 100% functional. I'm on a quest to train a head swap to make the process a bit easier. The problem is that so far nobody has trained ref2va models, only fl2va. To make this work, I'm having to train an adapter to do cfg de-distill and only after that be able to train the head swap.
DiffSynth now provides a DeCFG training adapter. I'm currently training an H3 head-swap LoRA, but the main challenge is that I don't have high-quality paired source-target data.
DiffSynth now provides a DeCFG training adapter. I'm currently training an H3 head-swap LoRA, but the main challenge is that I don't have high-quality paired source-target data.
I've already trained one, a really good one actually, I just haven't released it yet.
video must be processed to make the success rate increased, looking forward to your h3 version😄