MiniMax-H3 INT8 ConvRot

简体中文

Converted MiniMax-H3 weights for ComfyUI-RH-MiniMax-H3. This repository is a complete ComfyUI bundle: INT8 ConvRot DiT / text-encoder weights, merged VAEs, FL2VA / Ref2VA configs, and optional turbo LoRAs.

The same package is mirrored on ModelScope. Download from one source. The full snapshot is about 99 GiB; keep at least 120 GiB free.

Run Online and API Access

If you do not want to download the full 99 GiB model bundle locally, you can try MiniMax-H3 related workflows online on RunningHub, or integrate RunningHub AI app / workflow APIs into your own product.

The online demo is designed for users who want to quickly experience MiniMax-H3 video and audio generation without setting up a local GPU environment. You can first preview and run the public model on RunningHub, then choose to download the full weights locally or connect similar workflows to your product through RunningHub API.

Want to try it instantly? Open the MiniMax-H3 RunningHub AI Generator and run it online. Want to integrate video / audio generation into your product? Start with the RunningHub API Documentation.

RunningHub API supports model APIs, AI app APIs, workflow APIs, and LLM APIs. For MiniMax-H3 video / audio generation workflows, the recommended task flow is:

Submit task -> get taskId -> check task status -> retrieve generated result

Recommended use cases:

Run MiniMax-H3 video / audio generation workflows online
Add text-to-video, image-to-video, or video-to-audio generation to your product
Provide creators, developers, and teams with no-local-setup generative AI workflows
Build automated video generation, content production, or multimodal applications with RunningHub API

File	Size	Role
`MiniMax-H3-FL2VA-int8_convrot.safetensors`	31.65 GiB	FL2VA / T2VA / V2A DiT
`MiniMax-H3-Ref2VA-int8_convrot.safetensors`	31.65 GiB	Ref2VA DiT
`qwen3-vl-32b-int8_convrot.safetensors`	25.28 GiB	Qwen3-VL text encoder
`MiniMax-H3-video_vae.safetensors`	4.85 GiB	Video VAE
`MiniMax-H3-audio_vae.safetensors`	0.56 GiB	Audio VAE
`minimax_h3_fl2v_turbo_4step_v0.1.safetensors`	1.29 GiB	Optional FL2VA / T2VA 4-step turbo
`minimax_h3_fl2v_turbo_4step_v1.0_768p_bf16.safetensors`	1.29 GiB	Optional FL2VA / T2VA 4-step turbo for 768p
`minimax_h3_fl2v_turbo_8step_v1.0_bf16.safetensors`	1.29 GiB	Optional FL2VA / T2VA 8-step turbo
`minimax_h3_ref2v_turbo_4step_v0.1_bf16.safetensors`	1.29 GiB	Optional Ref2VA 4-step turbo
`FL2VA/`	configs	FL2VA tokenizer, processor, and component configs
`Ref2VA/`	configs	Ref2VA tokenizer, processor, and component configs

Install into ComfyUI

ComfyUI/
└── models/
    └── MiniMax-H3-INT8-CONVROT/
        ├── MiniMax-H3-FL2VA-int8_convrot.safetensors
        ├── MiniMax-H3-Ref2VA-int8_convrot.safetensors
        ├── qwen3-vl-32b-int8_convrot.safetensors
        ├── MiniMax-H3-video_vae.safetensors
        ├── MiniMax-H3-audio_vae.safetensors
        ├── minimax_h3_fl2v_turbo_4step_v0.1.safetensors
        ├── minimax_h3_fl2v_turbo_4step_v1.0_768p_bf16.safetensors
        ├── minimax_h3_fl2v_turbo_8step_v1.0_bf16.safetensors
        ├── minimax_h3_ref2v_turbo_4step_v0.1_bf16.safetensors
        ├── FL2VA/
        └── Ref2VA/

cd /path/to/ComfyUI
python3 -m pip install -U huggingface_hub
hf download Gluttony10/MiniMax-H3-INT8-CONVROT \
  --local-dir ./models/MiniMax-H3-INT8-CONVROT

China users can use the ModelScope mirror:

cd /path/to/ComfyUI
python3 -m pip install -U modelscope
modelscope download --model Gluttony10/MiniMax-H3-INT8-CONVROT \
  --local_dir ./models/MiniMax-H3-INT8-CONVROT

Re-run the same command to resume or update an interrupted download. On a high-bandwidth machine with at least 64 GiB RAM, prefix hf download with HF_XET_HIGH_PERFORMANCE=1.

The plugin auto-detects this complete converted bundle. The legacy ComfyUI/models/MiniMax-H3 directory remains supported.

Optional Turbo LoRA

Select one turbo LoRA in the model loader for fewer sampling steps. Leave the LoRA empty to run the base converted weights.

File	Typical use
`minimax_h3_fl2v_turbo_4step_v0.1.safetensors`	FL2VA / T2VA 4-step turbo
`minimax_h3_fl2v_turbo_4step_v1.0_768p_bf16.safetensors`	FL2VA / T2VA 4-step turbo for 768p
`minimax_h3_fl2v_turbo_8step_v1.0_bf16.safetensors`	FL2VA / T2VA 8-step turbo
`minimax_h3_ref2v_turbo_4step_v0.1_bf16.safetensors`	Ref2VA 4-step turbo

Plugin

Install ComfyUI-RH-MiniMax-H3 and use the nodes under RunningHub/MiniMax H3:

Node	Purpose
`RHMiniMaxH3ModelLoader`	Load the FL2VA / Ref2VA DiT and optional LoRA
`RHMiniMaxH3TextEncoderLoader`	Load the Qwen3-VL text encoder
`RHMiniMaxH3VAELoader`	Load the video and audio VAEs
`RHMiniMaxH3VideoGen`	T2VA / FL2VA / V2A
`RHMiniMaxH3RefGen`	Ref2VA with ordered image, video, and audio references

Example workflows:

License

Converted weights follow the upstream MiniMax-H3 Community License. The ComfyUI plugin code is Apache-2.0.

Model tree for RunningHubAI/MiniMax-H3-INT8-CONVROT

Base model

MiniMaxAI/MiniMax-H3

Finetuned

(76)

this model

RunningHubAI
/

MiniMax-H3-INT8-CONVROT