MiniMax H3 Turbo GGUF

MiniMax H3 Turbo is a distilled FL2VA variant that runs in 4 steps at 768p. It generates video with native stereo audio from text, one frame, or first and last frames. The GGUF files hold the base model plus distilled weights merged into one checkpoint.

Hugging Face repositoryMiniMax-H3-Turbo-GGUF on Hugging Face

Example output

Available Quantizations

Quant	Size	File
Q4_0	11.4 GB	minimax_h3_fl2v_turbo_4step_v1.0_768p_Q4_0.gguf
Q8_0	21.4 GB	minimax_h3_fl2v_turbo_4step_v1.0_768p_Q8_0.gguf
Q8_CR	20.2 GB	minimax_h3_fl2v_turbo_4step_v1.0_768p_Q8_CR.gguf

Model Input and Output

Variants and inputs

Variant	Inputs
Text-to-Video	A text prompt only.
Image-to-Video	A text prompt and one frame on first_frame.
First and Last Frame to Video	A text prompt and two images on first_frame and last_frame.

Outputs

Field	Value
Steps	4 NFE
Resolution	768p (1344 x 768)
Frame rate	24 FPS
Audio	32 kHz stereo
Video shift	6
Audio shift	3

ComfyUI Setup

Load the model with the GGUF loader node from comfyui-gguf-reboot. The loader node gives the best performance. You also need the MiniMax H3 Turbo sampler nodes, from comfyui-minimax-h3-turbo on the Comfy registry.

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 workflow