lightx2v

/

Minimax-h3-Turbo

Like 966

Thank you very much!

I would like to know what the training setting difference between v1.0 and v.1.1?

I try to train my own 4nfe fl2va lora under the released training code but find that:

1. the motion, texture and layout all look good, but
2. colors always become warmer than 20 nfe baseline which tend to be cooler and darker.

This issue happens to inference with the released minimax_h3_fl2v_turbo_4step_v1.0_768p lora ckpt as well.
But I find minimax_h3_fl2v_turbo_4step_v1.1_768p fixed the color bias! That's amazing!

So I want to know what the difference between v1.0 and v1.1 during training ? any hypeparameter change, different training data ?

Thank you very much!