What is this used for?【minimax_h3_ref_lora_rank_256_bf16】
As the title says
i think fl2va+ref_lora = ref2va 🤔 but not sure at what strength the lora should be used.
I tried it. At strength 1 my video character looks more identical to my photo reference. I think it's a LORA that makes your character looks much closer to the reference photo. I could be wrong, just guessing from what I saw from my generated videos using this LORA.
There is potentially a very powerful use case: Hybrid Conditioning.
https://github.com/kitsune123150/minimax-h3-hybrid-cond
From my experience character consistency with I2V is poor but R2V is excellent. What better to have it start from a defined start frame.
I would say that may be more met by a checkpoint merge though. But either the I2V or R2V checkpoints seem to be capable of the other's capabilities to some extent; actually, for a while I was using the I2V checkpoint with the R2V node and didn't realize it.
Is this lora loaded on a standard node or on a turbo lora node?