lightx2v

/

Minimax-h3-Turbo

like 494

much better after using minimax_h3_fl2v_lightx2v_turbo_4step_v0.1_comfy_resized_avg_rank_21_bf16.safetensors
we need a v1.0 version resized avg rank 21 lora

https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI/blob/main/minimax_h3_fl2v_turbo_4step_v1.0_768p_comfyui_resized_avg_rank_21_bf16.safetensors

I know some will say oh motion is fine but trust me its not, all the lora makers right now have the same issue the motion is being killed, having said that if you download minimax_h3_fl2v_lightx2v_turbo_4step_v0.1_comfy_resized_avg_rank_21_bf16.safetensors by kajai and run that in addition to this lora the motion fluidity returns. I'm not sure what he's doing but that file fixes that crap motion.

I know some will say oh motion is fine but trust me its not, all the lora makers right now have the same issue the motion is being killed, having said that if you download minimax_h3_fl2v_lightx2v_turbo_4step_v0.1_comfy_resized_avg_rank_21_bf16.safetensors by kajai and run that in addition to this lora the motion fluidity returns. I'm not sure what he's doing but that file fixes that crap motion.

i believe you, i've noticed the same thing, but could you tell me at what strengths do you use them when you mix them to get the best results?

https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI/blob/main/minimax_h3_fl2v_turbo_4step_v1.0_768p_comfyui_resized_avg_rank_21_bf16.safetensors
A new issue has arisen: using minimax_h3_fl2v_turbo_4step_v1.0_768p_comfyui_resized_avg_rank_21_bf16.safetensors results in abnormal audio generation, whereas using minimax_h3_fl2v_lightx2v_turbo_4step_v0.1_comfy_resized_avg_rank_21_bf16.safetensors causes changes in how the generated video adheres to the prompts.

i believe you, i've noticed the same thing, but could you tell me at what strengths do you use them when you mix them to get the best results?

Before with kaijis version i used to just limit the main lighting lora but now with this one it looks like we have to limit the rank lora or we get disney style over the top animation, im still trying to finetune whats perfect but starting at 0.3- 0.5 for str for the rank lora is a good starting point

I know some will say oh motion is fine but trust me its not, all the lora makers right now have the same issue the motion is being killed, having said that if you download minimax_h3_fl2v_lightx2v_turbo_4step_v0.1_comfy_resized_avg_rank_21_bf16.safetensors by kajai and run that in addition to this lora the motion fluidity returns. I'm not sure what he's doing but that file fixes that crap motion.

you mean run two turbo lora to fix motion? What is the lora strength