Zhenzhen MiniMax Hailuo Video

If you've envied Hailuo (MiniMax) video results but don't want to babysit the MiniMax website, this is the node that gets you there from inside ComfyUI. Like everything in the Comfly pack it's a paid API client - your prompt goes out to MiniMax through the Comfly relay, a render happens on their hardware, and minutes later a video comes back. The README is explicit that this one covers "all of Hailuo's video models," including first/last-frame support, so it's less a toy and more the whole Hailuo catalog behind one node.

The model dropdown is where the real differences live, so don't just leave the default:

• MiniMax-Hailuo-02 - the current flagship. Default, and a fine default.
• T2V-01 - text-to-video workhorse.
• I2V-01 / I2V-01-Director - image-to-video; feed first_frame_image. The Director variants add camera-control prompting.
• I2V-01-live - the "moving selfie" style generation.
• S2V-01 - subject-to-video: you feed subject_reference and it keeps that character consistent across shots.

The author points you at MiniMax's official docs for exact per-model capabilities, which is the right move - model features change faster than this article ever will.

Inputs worth touching

• prompt - multiline, required.
• duration - 6 or 10 seconds.
• resolution - 720P, 768P (default), or 1080P.
• prompt_optimizer - default on; it rewrites/expands your prompt before submission. Leave it on unless your prompt is already a well-tuned paragraph, because it will happily "improve" things you meant to keep literal.
• fast_pretreatment - off by default; when on it's the quicker preprocessing path. Fine for drafts.
• first_frame_image / last_frame_image - for I2V and first/last frame work.
• subject_reference - the one that matters for S2V-01 character work.
• seed - set it to reproduce a shot, leave 0 for variety.

Outputs are video (wire it straight into ComfyUI's native Save Video - that's what the README recommends), plus task_id and response for debugging.

Install and key

Shared with the whole pack:

cd ComfyUI/custom_nodes
git clone https://github.com/ainewsto/Comfyui_Comfly

Or via ComfyUI Manager, search "Comfyui_Comfly", restart. No model files, no VRAM - it's an HTTP client. Create an account at https://ai.comfly.chat, drop the key into Comflyapi.json (or the Comfly_api_set node), and you're done; the api_key input on the node is just a per-run override.

Gotchas

Model choice silently changes which inputs are honored - S2V-01 without subject_reference is a waste of a paid call, and I2V without a first frame is just a worse text-to-video. Check the MiniMax docs before you pick. Rendering takes a while and it costs per run, so iterate cheap (768P, 6s) and only go 1080P/10s when you've locked the shot. And remember: everything here runs on a third-party relay, so a server hiccup just fails the task - Comfly refunds failed async jobs, and the hk/us fallback lines in Comfly_api_set are your outage escape hatch.