MiniMax CN T2V Config

Hailuo is MiniMax's video model family, and it's one of the more talked-about hosted video APIs out there. MiniMaxCNT2VConfig is the node that turns your text prompt into a Hailuo text-to-video request inside ComfyUI - no GPU, no Wan checkpoint to babysit, no VACE, just a prompt and a key.

It's a config node, so it emits a MEDIAHUB_VIDEO_REQUEST that you feed into MediaHubSubmitVideo, then MediaHubResolveVideo, then SaveVideo. The chain is identical for every video provider in this pack; only the config node changes.

Inputs that matter

• model - four choices: MiniMax-Hailuo-2.3 (the current quality model), MiniMax-Hailuo-02, and the T2V-01 / T2V-01-Director pair. The Director models are MiniMax's nod to camera-direction prompting - you can ask for dolly shots and crane moves in the prompt and expect them to actually land.
• prompt - your script. With Director models, describe the shot, not just the scene.
• resolution - 768P or 1080P. Simple, and 1080P costs more.
• duration - 6 or 10 seconds. That's the whole choice; no fine-grained frame control here, unlike the Agnes video nodes.
• prompt_optimizer - default on for video (opposite of the image node). MiniMax rewrites your prompt before generating. Leave it on; it's usually the difference between "okay" and "actually good" for video, and you can see the effect in raw_json.
• fast_pretreatment - off by default; a speed-up pre-pass that trades a bit of quality for faster turnaround. Flip it on for drafts.
• aigc_watermark - off by default.

Output: request → MediaHubSubmitVideo.

Why you'd reach for it

Hailuo quality at 10 seconds with zero local resources is a strong pitch, and the Director models are genuinely nice for shot-planning prompts. If you're comparing hosted video APIs, this node plus the Seedance and Wan ones in the same pack give you an apples-to-apples way to run the same prompt through each provider and judge for yourself.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/vantang/ComfyUI-MediaHub.git
/path/to/ComfyUI/python -m pip install -r ComfyUI-MediaHub/requirements.txt

Restart ComfyUI; node under MediaHub/Provider/MiniMax CN/Video. Only requests/Pillow/numpy - renders happen at api.minimaxi.com, so you need a MiniMax CN key, and remember that's the mainland account, not MiniMax Global.

Common gotchas

Model availability is the one that bites: not every model is available to every account/region, and a 403 or model-not-found error usually means access rather than a typo - check raw_json before hunting for a node bug. Also remember the whole pipeline: config → Submit → Resolve. The resolve step polls up to ten minutes, so a 10-second 1080P render can genuinely sit in the queue for a while; that's normal. And if prompt_optimizer is mangling a prompt you carefully wrote, that's exactly what it's for - turn it off if you need your words to reach the model untouched.