ComfyUI Usage & Node Requirements
To run these GGUF models in ComfyUI, install the following custom node packages:
GGUF Loader Node :
ComfyUI-GGUF (GitHub) or install via ComfyUI Node Registry: comfyui-gguf-reboot
Used to load the quantized
.ggufUNet/diffusion models.
MiniMax H3 Turbo Sampler:
ComfyUI-MiniMax-H3-Turbo (GitHub) or install via ComfyUI Node Registry: comfyui-minimax-h3-turbo
Provides the dedicated Turbo sampling schedules and shift calculations for 4-step generation.
Model Placement
Place the downloaded GGUF files in your ComfyUI models directory:
ComfyUI/models/diffusion_models/minimax_h3_fl2v_turbo_4step_v1.0_768p_Q8_CR.ggufShift & Sampling Schedule
Recommended inference step count: 4 steps.
Minimax H3 Video shift: 6, Audio shift: 3
Minimax H3 Turbo Sampler
simple scheduler
Supported Input Modes (FL2VA)
Text-to-Audio-Video (T2VA): No image input connected.
Image-to-Audio-Video (I2VA): One image connected to
first_frame.First-and-Last-Frame-to-Audio-Video (FL2VA): Two images connected to
first_frameandlast_frame.
ComfyUI Workflows
Example workflows are provided
workflows/minimax_h3_t2v-turbo-gguf.json— Text-to-Audio-Video (T2VA) Turbo workflow.workflows/minimax_h3_i2v-turbo-gguf.json— Image-to-Audio-Video (I2VA / FL2VA) Turbo workflow.
License
MiniMax H3 is released under the MiniMax H3 Community License Agreement.
Description
FAQ
Comments (6)
As result quality is Better the INT4 or the GUFF Q4 someone have tried a bit of comparison?
At the moment tryed a bit the INT4 but I'm getting better result with WAN 2.2
I use INT8 GGUF (Q8_CR) and it's fine, but I think steps count makes more difference than quant levels. Q4 GGUF and Q8_0 GGUF are also fine, I have not checked other formats
@molbal Thanks for the feedback. I need to find the time to tinker a bit, ITN8 crash badly on my PC for now with 16GB of VRAM some optimization or sacrifice in length/resolution will be probably needed.
@ai_phanotm_red1893 I'm doing some research with INT4, I'll make sure to ping you if I reach something usable (It could go either way now)
@molbal Thanks. In the meantime I was able to get INT8 working for 8s clip. Now working on getting some good quality out of it. AI move so fast that keeping up to new model is a full time job, very difficult to fit it in WE freetime :D
@ai_phanotm_red1893 Glad to hear it works :D Yes definitely, difficult to keep up with the news