Hailuo H3 视频生成

Hailuo H3 is the flagship tier of MiniMax's video line, and this node front-ends six of its variants: domestic and global T2V, I2V (first/last frame), and Multi - the mode that makes H3 worth the upgrade. Multi lets you throw up to nine reference images, three reference videos and three reference audios at one generation, and then reference them from the prompt as @Image 1, @Video 1, @Audio 1. You're building a scene from a mood board inside a single node, which is not something the pack's older video nodes can do.

It's another API wrapper from T8mars/ComfyUI_Seedance - key from api.seedance.nz, zero local compute.

How it works

Standard async video flow under the hood: the node collects whatever media the selected model accepts, uploads local IMAGE/VIDEO/AUDIO references, maps them to the API's image/video/audio fields, submits, polls, and streams the MP4 back down with a header check. The front end is what does the heavy lifting here - switch model and the image/video/audio slots appear or vanish to match what that mode accepts, so you can't casually wire a video into T2V.

The inputs that matter

• model - six choices: hailuo-h3-t2v, hailuo-h3-i2v, hailuo-h3-multi, and their -global- twins.
• prompt - required for T2V and Multi (Multi prompts use the @Image 1 / @Video 1 / @Audio 1 syntax), optional for I2V.
• seconds - 5 to 15.
• resolution - 768P or 2K. H3 is a 2K-class model, which is its whole point over Hailuo 2.3.
• ratio - used by T2V and Multi (including adaptive); I2V follows the input frame and doesn't send a ratio at all.
• image1..image9, video1..video3, audio1..audio3 - the Multi reference pool. I2V uses image1 as the first frame and image2 as the optional last frame.

Outputs: video, video_url, task_id, response.

Installing it

Same as every node in this pack. ComfyUI Manager → search "ComfyUI Seedance", or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git
cd ../..
python -m pip install -r custom_nodes/ComfyUI_Seedance/requirements.txt

Restart, key from api.seedance.nz/console, Seedance API Config node into api_config.

Common issues

• References not being used - check the mode. Multi is the only one that reads videos and audios; T2V takes no reference material at all.
• @Video 1 reference ignored - the prompt references only work meaningfully in Multi mode, and the tooltip is explicit that Multi accepts up to 3 videos and 3 audios. Exceeding that, the extras are dropped.
• Slow upload phase - up to 9 images + 3 videos + 3 audios is a lot of uploads, and the gateway rate-limits uploads per token. The node waits and retries; a long "uploading" phase before generation is normal, not a hang.

The honest framing, as with all closed-model nodes: your references leave the machine and land on a gateway you don't control, subject to the model's own content policy. For a model this good that can't run locally, that's the trade - and Multi is exactly the kind of job worth paying it for.