How to Generate Short Video Clips With MiniMax Hailuo
Hailuo is MiniMax’s text-to-video and image-to-video model, accessible through the Hailuo AI web app and through MiniMax’s API. Clips come out at 6 or 10 seconds depending on the mode, in 720p or 1080p, with a first-frame image option that most people should be using.
It is not a film tool. It is a b-roll tool: short, loopable, visually consistent snippets that fill gaps in a longer edit.
What does a Hailuo prompt need?
Four things, in this order: subject, action, camera, look. Prompts that skip the camera line get a locked-off shot whether or not that is what you wanted.
A workable example: “A barista pours milk into a ceramic cup, close-up on hands, slow push in, soft morning window light, shallow depth of field.” That is 23 words and it produces something usable most attempts.
Where prompts go wrong is stacking adjectives. “Cinematic stunning ultra-detailed 8K hyper-realistic” adds nothing the model can act on. Concrete lighting and lens language does.
How many attempts does a usable clip take?
Budget three generations per usable clip. For people or hands, budget five. The model handles landscapes, vehicles, and abstract motion well; it still struggles with fingers wrapping around objects and with crowds maintaining consistent faces.
If a generation produces the right motion with a broken detail in frame, try the image-to-video path: take a still that is correct and animate it. That reduces the variance a lot.
What are the resolution and duration limits?
| Mode | Duration | Resolution | Notes |
|---|---|---|---|
| Text to video | 6s / 10s | 720p, 1080p | 10s uses more credits |
| Image to video | 6s / 10s | 720p, 1080p | Accepts a first-frame image |
| API access | Programmable | Up to 1080p | Priced per generation |
Exact credit costs change; check the current pricing page before committing a batch. The API is the cheaper route for anything over a few dozen clips a month.
How do you stitch clips into something watchable
Six-second clips do not edit themselves. Practical approach:
- Generate 8 to 12 clips around one theme, not 12 unrelated shots.
- Keep the aspect ratio consistent across the set.
- Use a single colour reference in the prompt (“warm amber light”) so the cuts feel like one shoot.
- Cut on motion — end each clip mid-move so the transition is hidden.
For teams building social cuts, Pictory Video is a sensible assembly layer: it accepts generated clips and handles captions and pacing. If you need a real presenter instead, Tella (AI Recorder) records screen and camera with automatic trimming.
Who should use it, and who should not
Use it if you need b-roll for content you already scripted, product atmosphere shots, or background footage for a launch page. The cost per usable second is low compared with licensing stock.
Do not use it for anything requiring brand-accurate text, real people’s likenesses, or a continuous narrative longer than about 10 seconds. Text in generated footage is garbled nearly every time, and a shot that must match a competitor’s product will not.
One practical warning: MiniMax Hailuo generations are slow at peak times. Queue a batch overnight rather than in the hour before a deadline.
Quick Summary
- Hailuo generates 6s or 10s clips at up to 1080p from text or a still image.
- Prompt order: subject, action, camera, look. Skip filler adjectives.
- Budget three generations per usable clip; five for hands and crowds.
- Image-to-video reduces variance when you already have a good frame.
- No reliable on-screen text, and peak-hour queues are long.
