MiniMax H3
2K omni-modal video, references and all, via Civitai
- first_frame_image
- last_frame_image
- reference_images
- api_config
- video
- draft_cache_audio
- draft_cache_video
- workflow_id
- raw_json
MiniMax H3 is the 33B omni-modal video model that finally gave the open-weights world a real answer to Veo's native-audio trick - text, image, video and audio handled as one input context, generating clips up to 2K with stereo sound. There's a catch baked into the local release: the community license geofences out the US, EU, UK and Korea, so if you live in one of those places you're not licensed to run the weights yourself. Which is exactly why this node is interesting - it's H3 running on Civitai's cloud, where the licensing calculus is Civitai's problem, not yours. Pay in Buzz, get 2K video back.
It's part of civitai-comfy-nodes, Civitai's official ~160-node pack generated from its Orchestration API spec. Cloud job: submit a videoGen tagged for MiniMax H3, long-poll, get a native VIDEO tensor. This is the full-cloud variant - the menu also has two MiniMax H3 (Comfy) nodes that expose a Comfy-style parameter set (width/height/steps/turbo); those are the tinkerer's version, this one is the API version.
Inputs that matter
Three required, which is unusual for this pack:
- prompt - the content.
- aspect_ratio -
adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16. - duration - 5 to 15 seconds.
- resolution - a single choice:
2K. H3's headline resolution, and there's no cheaper tier - so duration is your only cost lever.
Then the reference stack, all optional:
- first_frame_image / last_frame_image (IMAGE sockets) - pin the bookends.
- reference_images (IMAGE) - character/subject/style references.
- reference_videos_json / reference_audios_json (STRING) - reference clips as JSON, for motion or voice/style transfer. Advanced.
- watermark - off by default; flip on if you want the output marked.
- reference_video_seconds - auto-filled during initialization; per the tooltip, MiniMax bills uploaded reference video at the same per-second rate as output, so this has to be known before the cost estimate is charged.
Outputs: video (VIDEO), draft_cache_audio (AUDIO), draft_cache_video (VIDEO), workflow_id and raw_json (STRING).
Install and auth
Standard for the pack - official, Comfy Registry:
# ComfyUI Manager: search "Civitai Comfy Nodes"
cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt
Restart, authenticate via a Civitai Auth node, CIVITAI_API_TOKEN, or sidebar OAuth.
Common issues
2K-only means cost scales with duration and retries - there's no draft resolution to fall back on, so iterate on the prompt at the minimum duration. Reference video bills at output rates - that reference_video_seconds note is the model being honest about the meter; a long reference clip is part of your bill. The reference JSON fields are the fiddly part - no editor, paste valid JSON or the job rejects. Your images, videos, audio and prompt all leave your machine to a third-party server. And the pack is early preview - behavior can change without notice.
Inputs (12)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| aspect_ratio | COMBO | 7 options: adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, +1 | |
| duration | INT | 55–15 | — |
| resolution | COMBO | 1 options: 2K | |
| first_frame_imageopt | IMAGE | Either A URL, A DataURL or a Base64 string | |
| last_frame_imageopt | IMAGE | Either A URL, A DataURL or a Base64 string | |
| reference_imagesopt | IMAGE | — | |
| reference_videos_jsonopt | STRING | — | |
| reference_audios_jsonopt | STRING | — | |
| watermarkopt | BOOLEAN | false | — |
| reference_video_secondsopt | INT | 00–2147483647 | Measured total duration of Civitai.Orchestration.Grains.Workflows.Steps.VideoGen.MiniMaxH3VideoGenInput.ReferenceVideos, filled in during initialization. MiniMax bills uploaded reference video at the same per-second rate as output, so this has to be known before the estimate is charged. |
| api_configopt | CIVITAI_CONFIG | Optional Civitai Auth connection; defaults to CIVITAI_API_TOKEN or stored OAuth login. |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| video | VIDEO | — |
| draft_cache_audio | AUDIO | — |
| draft_cache_video | VIDEO | — |
| workflow_id | STRING | — |
| raw_json | STRING | — |