Ref2‑VA + FL2V 4‑step turbo LoRA causes artifacts on fight scenes, request for native Ref2‑4step LoRA
#13
by aronzhan - opened
I'm forcing the combination of minimax_h3_ref2va_int8_convrot model with minimax_h3_fl2v_lightx2v_turbo_4step_v0.1_comfy.safetensors.
It works perfectly for dialogue‑driven narrative videos.
However, fight‑scene outputs suffer from severe artifacts, body distortion and blurring.
When will you release the Ref2‑compatible 4‑step LoRA?