MiniMax H3 Open Weights in ComfyUI: Native Audio-Video T2V, I2V and R2V in Three Official Workflows

MiniMax H3 has released its weights. It is not the first native audio-video model — LTX-2.3 already supported joint audio-video generation — but the author argues it lands in the first tier of open-weight audio-video models, and community reaction has been strong. The notes below come from the episode subtitles and the published mirror page; no workflow JSON was shipped or parsed, so no node-level parameters are claimed beyond what the episode states.

What the mirror provides

The model runtime and workflows are packaged into a one-click image with three official ComfyUI workflows:

• T2V — text to audio-video.
• I2V — image to audio-video.
• R2V — reference to video.

The author states the three workflows come directly from the official release; the only work done here is the online deployment.

Text-to-video details from the episode

• No image upload needed; just edit the prompt box in the middle.
• Write character, scene, action, camera language, environmental sound, SFX and background music in the order they occur. The example: a person running, camera tracking from the side, rain around, cinematic sound effects.
• On the left there is a resolution selector: aspect ratio and resolution. The author uses the official defaults — 16:9, pixel tier 0.4 → 864×480, suggested duration 5 seconds.
• Width and height should be multiples of 32.
• Start from defaults, then raise settings until you hit the performance ceiling.

The author also points to a prompt skill placed in the Bilibili description that can be installed or fed to an LLM to speed up prompt writing.

Hardware

The author recommends 50-series or above, with a 5090 offering a good price-performance experience.

Deploying the mirror

1. Open the mirror page: https://www.compshare.cn/images/eciFeWNe3kJk?referral_code=A7fTQUYY9lCDFJwA6MMvoq
2. Deploy an instance and wait for the image to come up.
3. Open ComfyUI, pick T2V, I2V or R2V, and run.

Links

• Mirror (with referral code): https://www.compshare.cn/images/eciFeWNe3kJk?referral_code=A7fTQUYY9lCDFJwA6MMvoq
• Video tutorial (YouTube): https://www.youtube.com/watch?v=wedGmLy63tQ
• Chinese original (Bilibili): https://www.bilibili.com/video/BV1AHuF6hEH2

Limits

• Default settings (16:9, 0.4 pixel tier, 5 seconds) are the episode's starting point, not a universal optimum.
• The three workflows are official; the deployment is the author's.
• The mirror is a deployment convenience, not a quality guarantee.