Recommended inference settings for FL2V Turbo 4-step v1.1 (768p)
We have released the updated 768p ComfyUI BF16 checkpoint:
minimax_h3_fl2v_turbo_4step_v1.1_768p_comfyui_bf16.safetensors
For optimal generation quality, we recommend using the following inference settings:
shift_video:6.00shift_audio:3.00steps:4sampler_name:euler
These settings are specifically recommended for this v1.1 checkpoint.
scheduler?
scheduler?
Just use simple
Thanks, going to try. What is this about btw?
When generating video at 1MP resolution and 4 steps, the audio still gets corrupted, and ghosting artifacts persist.
thanks Again TEAM!! but what node is this?
thanks Again TEAM!! but what node is this?
Subgraph
I have now done some tests with the new Lora, first of all thank you for the effort and that it is being made available for free. But I'm having trouble getting a sharp picture. I've tried exactly the settings, and the faces and especially the people further away look like they're in a horror movie. I went through various models fp8, bf16, int8, pruned and unpruned models, various shifts and sigmas, samplers etc., 1mp resolution. I just can't seem to get good results with the lora, which I find very unfortunate. Have you by any chance found any settings that work well? Best regards
Everything is good, strength 1 of the lora was not enough, I now have very good results with 1.20-1.30.
The results with the higher strength are insane, thank you for the great lora!
Everything is good, strength 1 of the lora was not enough, I now have very good results with 1.20-1.30.
The results with the higher strength are insane, thank you for the great lora!
Thank you for the tip! With 1.30 it is almost perfect! I'm using it with 6 Steps at the moment and I'm really happy with that. This version seems to work waaaaay better than the old 1.0 version, at least for me.
what are these 2 shift fields?
my default workflow from comfy template doesn't have it...
what are these 2 shift fields?
my default workflow from comfy template doesn't have it...
MiniMaxH3SigmaShift node
what are these 2 shift fields?
my default workflow from comfy template doesn't have it...MiniMaxH3SigmaShift node
The ModelSamplingMinimaxH3 node? is it the same thing? why is it different for me? I've been using that node with these shifts.
what are these 2 shift fields?
my default workflow from comfy template doesn't have it...MiniMaxH3SigmaShift node
The ModelSamplingMinimaxH3 node? is it the same thing? why is it different for me? I've been using that node with these shifts.
Yes, your Comfy version is more uptodate. They have changed the name and behavior in earlier versions. SigmaShift was the old name.
I hope ref version coming soon ? using this makes the footage dark and high contrast
I hope ref version coming soon ? using this makes the footage dark and high contrast
Use your prompt to set natural tones. Don't rely on the LoRA alone. H3 benefits from camera settings in the prompt.