MiniMax Hailuo 02 Video
The text-to-video node with a deceptively long input list
- first_frame_image
- VIDEO
MiniMax Hailuo is the text-to-video workhorse of the closed video lineup, and this built-in node is how you use it without leaving ComfyUI. Its reputation among people who actually generate is "solid, middle-of-the-road, occasionally great" - it won't dethrone Veo on wild camera moves, but it's dependable, and Hailuo-02 handles a plain descriptive prompt better than most of the field. If the phrase "a person walks through a rain-soaked alley at dusk" needs to become footage, this node will do it.
Like every node in the built-in partner/video category, it's an API node: ships with ComfyUI core, no local models, calls MiniMax's API through Comfy Org's backend, and bills your Comfy account in credits. The task shows as "Queued / Processing" in the UI while Hailuo renders, then the finished clip arrives as a VIDEO output.
The inputs that matter
- prompt_text - the whole game. Hailuo is a text-to-video model, so this carries the scene, subject, and motion. Keep it descriptive and concrete.
- duration - 6 or 10 seconds, default 6. Here's the trap: at 1080p, Hailuo-02 only allows 6 seconds. Pick
10at1080Pand the node rejects the combination with a clear error. 10-second clips are a 768p thing. - resolution -
768P(1366x768) or1080P(1920x1080). Default 768P, and for Hailuo that's the sensible value - it's where the model is fastest and most stable. - first_frame_image - optional. Plug in an image and this becomes an image-to-video node instead; otherwise it's pure text-to-video.
- prompt_optimizer (default on) - MiniMax rewrites your prompt server-side for better results. Leave it on unless you've already hand-tuned your prompt and want it passed through verbatim.
- seed - the usual family behavior: a re-run toggle, non-deterministic results either way.
Output is a single VIDEO.
Where people get burned
- The 1080p duration lock. This is the one that surprises everyone. You will set 1080P, slide duration to 10, and get a validation error. The tooltip on
resolutionspells out the dimensions, and the model simply doesn't do 10s at full HD. - It's closed and it's billed. Hailuo sits in the "all closed source" column of every community roundup ("Vidu, Kling, Hailuo, all closed source"). No weights, no local version, per-clip pricing that climbs with resolution and duration. The badge is honest about it.
- Prompt optimizer can fight you. If you're prompting in a style you've already tuned, the automatic rewrite can drift it. Toggle it off for prompt fidelity; leave it on for convenience.
For a "prompt in, clip out, minimal fuss" node, this is one of the friendliest in the whole partner category. Just remember: 10 seconds lives at 768p, and every render is a line item.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt_text | STRING | Text prompt to guide the video generation. | |
| seedopt | INT | 00–18446744073709550000 | The random seed used for creating the noise. |
| first_frame_imageopt | IMAGE | Optional image to use as the first frame to generate a video. | |
| prompt_optimizeropt | BOOLEAN | true | Optimize prompt to improve generation quality when needed. |
| durationopt | COMBO | 6 | The length of the output video in seconds. |
| resolutionopt | COMBO | 768P | The dimensions of the video display. 1080p is 1920x1080, 768p is 1366x768. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| VIDEO | VIDEO | — |