Skip to content

Video

MiniMax Hailuo 02

Video

Available on Ofrai — chat, images, and video in one place with shared credits.

54 creditsModel details

Video studio

Preview

No clip yet

Your latest clip preview will show here after generation.

Open studio

Example prompts

Choose an example to load its prompt and recommended settings into the MiniMax Hailuo 02 creator above.

How it works

Create with MiniMax Hailuo 02 in three steps

  1. 01

    Choose your inputs

    Keep MiniMax Hailuo 02 selected, then add a prompt and any supported reference assets.

  2. 02

    Set the output

    Choose the aspect ratio, resolution, and available Video options before generating.

  3. 03

    Generate and refine

    Review the result, adjust the prompt or settings, and generate another version from the same place.

MiniMax Hailuo 02 is documented with one coupling that shapes almost every decision on it: the top resolution only accepts the short duration. 1080P is available at 6 seconds and nothing else, so 10-second clips exist only at 512P or 768P. That single rule decides which field you set first, and it is why the default of 768P is worth keeping while you iterate.

Beyond the coupling, this endpoint supports three generation modes — text-to-video, image-to-video and first-and-last-frame — and a camera-move system of 15 instructions that can either be combined inside one bracket or sequenced by their order in the prompt.

1080P locks you to 6 seconds

The resolution list is 512P, 768P and 1080P, with 768P as the documented default, and the duration list is 6 or 10 seconds with 6 as the default. The restriction that matters is the intersection: the reference states that 1080P supports only a 6-second duration. So a 10-second clip is a 512P or 768P clip, full stop, and asking for 1080P at 10 seconds is not a setting that will be quietly upgraded later — it is a request outside the documented range. Plan the field order accordingly.

Decide duration first from the edit, then pick the best resolution that duration allows, rather than choosing 1080P as a default and discovering at the end that your shot has to lose four seconds. If a deliverable genuinely needs both, the honest answer is two clips joined in the edit, which is usually preferable anyway because it gives you a cut point you can trim.

Three generation modes, and the prompt is only mandatory in one

The interface description lists text-to-video, image-to-video and first-and-last-frame generation, and the prompt field documents itself accordingly: it is required in text-to-video mode and optional in the image modes, where it can be used to describe the desired motion rather than the subject. That is a genuinely different authoring job in each mode. In text-to-video the prompt is doing all the work — subject, action, camera, style — inside a documented maximum of 2000 characters.

In the image modes the picture already carries the subject, and the prompt's job is to say what should happen to it. Writing a full subject description anyway, in an image mode, risks pulling the generation away from the image you supplied. Use the field for the motion you want: how the subject moves, how the camera travels, where the shot should settle.

Camera moves: combine inside a bracket, or sequence across the prompt

The endpoint supports 15 camera-move instructions, written in brackets, and the reference documents two ways of using them. Combination is several instructions inside one bracket, applying simultaneously — the documented example is a pan left combined with a rise — and the guidance is to keep it to three or fewer, which is a strong hint that stacking more produces interference rather than complexity.

Sequencing is ordering them by their position in the prompt text, so instructions appearing earlier take effect earlier; the documented pattern reads as push in, then pull out. The practical implication is that the brackets in your prompt are not decoration. They are the edit. Decide whether the shot moves as one composite gesture or as two distinct beats, write them accordingly, and resist the urge to add a third and fourth combination just because the list of 15 makes it possible. Available does not mean advisable.

Image requirements are checked before anything is generated

The first-frame image input has hard constraints, and they are the kind that is cheaper to satisfy at preparation time than to diagnose from a failed job. The accepted formats are JPG, JPEG, PNG and WebP. The file must be under 20MB. The short side must be greater than 300 pixels. The aspect ratio must fall between 2:5 and 5:2, which is a wide band — it admits both a tall 2:5 frame and a wide 5:2 frame — but it does exclude extreme strips.

When you pass exactly one image it becomes the starting frame of the video; when you pass nothing, the request is text-to-video. Note the field's own description covers the first-frame case, while the interface description also lists a first-and-last-frame mode, so if you need that mode, confirm which field carries the closing frame against the current reference rather than assuming the array is symmetrical.

What the 512P annotation does and does not tell you

The resolution list annotates the three values: 512P as standard definition and noted as available for image-to-video, 768P as high definition and the default, 1080P as full high definition at 6 seconds only. The 512P annotation is the one to read carefully, because it is phrased as a parenthetical rather than a restriction. The reference does not spell out further limits on 512P — it does not say the mode is unavailable for text-to-video, and it does not say what happens if you request it.

The safe reading is to treat 512P as the option to reach for when file size, upload time or draft volume matters more than image quality, and to validate it on a short test before committing a batch to it. Because the endpoint is asynchronous and returns a task ID immediately, a test render is cheap and the ambiguity resolves itself in one pass rather than in a post-mortem.

Planning a shot inside a 6-or-10-second envelope

Two duration values sounds restrictive until you notice what it implies about structure. At 6 seconds you have one gesture: an approach, a reveal, a single beat with a start and an end. At 10 seconds you have room for one gesture plus a settle, or for two camera moves in sequence, which is exactly where the ordered-instruction syntax pays off. Since 1080P is limited to 6 seconds, the highest-quality renders on this endpoint are also the ones with the least room, which changes how you use resolution.

Treat 1080P as the finishing pass for a shot whose 6-second shape is already settled, and use 768P as the working default while the camera work is still moving. If a scene needs two beats, plan them as two 6-second generations and cut between them, rather than asking a 10-second generation to earn a structure it has to invent.

FAQ

Can I get a 10-second video at 1080P?
No. The reference states that 1080P supports only a 6-second duration. 10-second clips are available at 512P or 768P, so choose duration first and then the best resolution that duration permits.
What are the requirements for the input image?
Format must be JPG, JPEG, PNG or WebP, size under 20MB, short side greater than 300 pixels, and an aspect ratio between 2:5 and 5:2. One image becomes the starting frame of the video; passing none makes it a text-to-video request.
How many camera moves should I combine in one bracket?
Keep combination brackets to three or fewer, which is what the reference recommends. Several instructions inside one bracket apply simultaneously, such as a pan left combined with a rise. To have moves happen one after another instead, place them in order in the prompt text.
Is the prompt required in image-to-video mode?
No. It is required for text-to-video and optional in the image modes, where it describes the motion you want rather than the subject, since the image already supplies that. Either way the documented maximum is 2000 characters.

FAQ

Can I use MiniMax Hailuo 02 directly on this page?
Yes. The creator above is the working Ofrai studio with MiniMax Hailuo 02 selected by default.
Where are my finished results saved?
Signed-in generations are kept in your Ofrai library so you can download or revisit them.
Which prompts work best with MiniMax Hailuo 02?
Describe the subject, setting, visual style, lighting, and camera or motion clearly. The examples above provide practical starting points.
Can I edit the example prompts before generating?
Yes. Load any example, then change its wording, aspect ratio, references, or other supported settings in the creator.

Compare other Video models that can handle similar creative work to MiniMax Hailuo 02.

Commercial use depends on the selected model and model terms.