MiniMax H3 GGUF

MiniMax H3 is an omni-modal generative system. It reads text, images, video, and audio and generates video with native stereo audio at up to 2K and 15 seconds. This repository ships GGUF quants of the FL2VA and Ref2VA variants.

Hugging Face repositoryMiniMax-H3-GGUF on Hugging Face

Example output

Available Quantizations

FL2VA

Quant	Size	File
Q4_0	11.4 GB	minimax_h3_fl2va_pruned_fp8_Q4_0.gguf
Q8_0	20.2 GB	minimax_h3_fl2va_pruned_fp8_Q8_0.gguf
Q8_CR	20.2 GB	minimax_h3_fl2va_pruned_fp8_Q8_CR.gguf
U16G	15.0 GB	minimax_h3_fl2va_pruned_fp8_U16G.gguf

Ref2VA

Quant	Size	File
Q4_0	11.4 GB	minimax-h3-ref2va-Q4_0.gguf
Q5_0	12.9 GB	minimax-h3-ref2va-Q5_0.gguf
Q8_0	21.4 GB	minimax-h3-ref2va-Q8_0.gguf
Q8_CR	20.2 GB	minimax-h3-ref2va-Q8_CR.gguf
U16G	15.0 GB	minimax-h3-ref2va-U16G.gguf

TIP

U16G targets 16GB cards. It selectively quantizes different layers to keep the model fit in 16GB VRAM. However, if your memory is fast, it might still be faster to use the Q8_CR quant.

Model Input and Output

Variants and inputs

Variant	Inputs
FL2VA	Zero, one, or two images. No image is text-to-video, one image is first or last frame to video, two images is first and last frame to video.
Ref2VA	Up to 9 images, up to 3 video clips, up to 3 audio clips. Audio must come with an image or video. Each clip is 2 to 15 seconds. Total across all inputs is 15 seconds and 12 files.

Outputs

Field	Value
Duration	4 to 15 seconds
Aspect ratios	21:9, 16:9, 4:3, 1:1, 3:4, 9:16
Resolution	768p default, 2K with H3-Regenerate-2K
Frame rate	24 FPS
Audio	32 kHz stereo
Language support	Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, Spanish

ComfyUI Setup

Load the model with the GGUF loader node from comfyui-gguf-reboot. The loader node gives the best performance.

Dependencies

Place these files in the ComfyUI model folders before running a workflow.

Component	File	Folder
Text encoder	qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors	models/clip/ or text_encoders/
Video VAE	minimax_h3_video_vae_fp16.safetensors	models/vae/
Audio VAE	minimax_h3_audio_vae_fp32.safetensors	models/vae/

Workflow downloads

Import these JSON files into ComfyUI.

JSON fileText-to-Video workflowJSON fileImage-to-Video workflowJSON fileReference-to-Video workflow