MiniMax H3: Open Omni-Modal Video Model With Native Audio

ComfyUI Wiki

MiniMax H3 is an open general-purpose omni-modal generation model: 768p video with native 32 kHz stereo audio, 11 languages, and 2K in-context regeneration.

M

MiniMax H3

Video GenerationNative AudioOmni-ModalText-to-VideoImage-to-Video

MiniMax H3 is an open general-purpose omni-modal generation system that understands multimodal context composed of text, images, video, and audio, and generates up to 15 seconds of video with native 32 kHz stereo audio in a single pass. It is the latest model in MiniMax's Hailuo video line, powered by a 33.1B dense single-stream omni transformer with a Qwen3-VL-32B text encoder.

H3 example input image from MiniMax's official demo

An example input image from MiniMax's official H3 multimodal context demo

MiniMax H3 was officially open-sourced on August 3, 2026 under the MiniMax H3 Community License Agreement. The open release covers H3-Base as two task-specific checkpoints: FL2VA (text-to-video and first/last-frame conditioning) and Ref2VA (reference-based generation). The H3-Context-IR preprocessing system and the H3-Regenerate-2K upscaling module remain hosted APIs.

Versions

VersionDescription
H3-Base FL2VAText-to-video, first-frame, last-frame, and first+last-frame modes
H3-Base Ref2VAReference-based generation from up to 9 images, 3 videos, and 3 audio clips

Key capabilities

  • Native stereo audio: 32 kHz stereo generated in the same pass as the picture, no separate audio model
  • Multimodal context: any combination of text, images, video, and audio as input
  • 2K in-context regeneration: H3-Regenerate-2K re-generates 768p output at 2K, reusing the original context
  • 11 languages: ar, zh, en, fr, de, it, ja, ko, pt, ru, es
  • Deployment: SGLang, vLLM, diffusers, and ComfyUI

Installation

MiniMax H3 is natively supported in ComfyUI. Place files in their respective folders:

FileDestination
minimax_h3_*_bf16.safetensors or *_int8_convrot.safetensorsComfyUI/models/diffusion_models/
qwen3vl_32b_minimax_h3_*.safetensorsComfyUI/models/text_encoders/
minimax_h3_video_vae_fp16.safetensorsComfyUI/models/vae/
minimax_h3_audio_vae_fp32.safetensorsComfyUI/models/vae/

The official ComfyUI repackaged repository (Comfy-Org/MiniMax-H3) provides bf16, INT8 convrot, and pruned INT8 diffusion model variants, plus bf16, INT8, and NVFP4 AWQ text encoder options. The pruned INT8 checkpoints are about 40% smaller thanks to precomputed adaLN curve tables, and the NVFP4 AWQ text encoder runs on any GPU.

Resources

Comments

Sign in with GitHub to join the discussion.

Loading comments…