MiniMax H3 Is Now Live on VIDEO AI ME

MiniMax H3 is now live in the VIDEO AI ME editor. You can generate with it today in three modes, text to video, image to video and reference to video, at 480p or 768p, in clips from five to fifteen seconds, with sound generated in the same pass as the picture. No install, no GPU, no waitlist.

This post covers what shipped, what the model is good at, and where it fits next to the other models already in the picker.

What is MiniMax H3?

MiniMax H3 is a general purpose video generation model from MiniMax, the team behind the Hailuo video products. It reads a written brief and produces a finished clip with synchronised audio: dialogue, ambient sound and effects, all in one generation rather than in a separate audio pass.

The practical difference from earlier generations is that H3 treats the whole brief as one context. Your subject description, your camera direction and your sound notes are read together, which is why a paragraph written like a shot list works better than a pile of comma separated keywords.

If you have used Hailuo before, our Hailuo review covers where that line started. H3 is the current generation of the same lineage. The model weights are published openly, and MiniMax hosts the model card on Hugging Face if you want the technical detail.

Why does MiniMax H3 matter for people who buy ads?

Because it is fast, and speed changes what you are willing to try.

When a draft takes a long time to come back, you protect your ideas. You write one careful prompt, you wait, and you talk yourself into whatever comes out because starting again is expensive. That is how teams end up running the first idea instead of the best one.

When drafts come back in seconds rather than minutes, the maths flips. You can write a hook, watch it, decide it is wrong, and rewrite it before your coffee is cold. Six versions of a hook stop being a project and become an afternoon. The cost of being wrong about a creative idea drops close to zero, so you stop defending ideas and start testing them.

That is the honest pitch for H3. It is not that no other model can produce a beautiful frame. It is that this one lets you find out whether an idea works before you have committed to it.

What can you generate with MiniMax H3 today?

Three modes, and you do not pick between them from a menu. The editor chooses based on how many reference images you attach.

Text to video, with no images attached

You describe the shot and the model builds it. This is the mode for hooks, concepts and any scene where you do not have footage or a product photo to anchor to. Aspect ratio is yours to choose here.

Image to video, with one image attached

The image you attach becomes the literal first frame and the model animates forward from it. This is the mode for product photos: you already have the pack shot, and you want it to move. The output aspect ratio follows your image, so crop before you upload.

Reference to video, with two to four images attached

The model carries the subjects, product and style from your reference images through a shot it composes itself. You address them in the prompt by their order, as Image 1, Image 2 and so on, which is how you keep a specific creator holding a specific product rather than getting a blend of both.

The H3 model itself accepts more than this, including reference video and reference audio clips. On VIDEO AI ME today you get text to video, image to video, and reference to video with up to four images.

Which settings does MiniMax H3 support?

Two notes that save you a wasted render. Image to video always follows the aspect ratio of the image you upload, so the aspect picker does not apply in that mode. And reference to video always renders at 768p, so if you attach two or more images the resolution choice is made for you.

We go deeper on the trade offs in our guides to durations, aspect ratios and the choice between 480p and 768p.

When should you pick MiniMax H3 over another model?

The picker has several models in it, and treating one of them as your favourite is how you end up using the wrong tool. The useful skill is choosing per job.

Reach for MiniMax H3 when the job is volume and iteration:

• Drafting hooks before you commit to one.
• Building the ten variants a paid social test actually needs.
• Turning a product photo into motion for a first pass.
• Any creative that will be watched on a phone at arm's length with sound on.

Reach for a slower model when the job is a single hero asset that has to be perfect, when you need a clip longer than fifteen seconds in one pass, or when a specific model handles your subject better. Seedance 2.5 goes to thirty seconds in a single generation, for example, which H3 does not.

We wrote a fuller head to head in MiniMax H3 against Seedance 2.5.

How do you start using MiniMax H3?

Open the editor, open the model dropdown, and select MiniMax H3. In the model picker it is listed as MiniMax H3 Max. Then set your duration, resolution and aspect ratio, write your prompt, and generate.

A first prompt worth copying, if you want to see what the model does with a straightforward ad brief:

A woman in her late twenties sits on a light grey sofa in a bright apartment, holding a small amber glass serum bottle up to the camera. She speaks directly to the lens with easy, unforced enthusiasm: "Three weeks. That is all it took." Soft natural window light from the left, shallow depth of field, handheld with a small amount of natural sway. Room tone only, no music. Vertical framing.

Generate that at 480p first. If the delivery lands, rerun it at 768p. If it does not, change one thing and go again, which is the whole point of a fast model.

For the full prompting method, including camera language and how to write the sound, see our MiniMax H3 prompt guide.

What does it cost?

Generation is billed in credits, and credits come with your plan. Starter includes 1,400 credits a month at $29, Pro includes 5,600 at $99, and Premium includes 12,000 at $199, with annual billing available on all three. Every plan carries the full commercial license, no watermarks, and unlimited projects.

Because rates move as models change, we do not publish a per second figure here. The pricing page always shows the current numbers, and the editor shows you the cost of a clip before you generate it.

Frequently Asked Questions

Is MiniMax H3 available on VIDEO AI ME right now?

Yes. MiniMax H3 is in the model picker in the editor, on any paid plan, in English, in the browser. There is nothing to install and no waitlist.

What clip lengths does MiniMax H3 support?

Five, eight, ten, twelve and fifteen seconds. Every length is generated in one pass rather than stitched together from shorter pieces.

Does MiniMax H3 generate sound?

Yes. Speech, ambient sound and effects are produced in the same pass as the picture, so what you describe in the prompt is what you hear in the file.

How many reference images can I attach?

Up to four. Attaching none gives you text to video, one gives you image to video, and two to four switch the shot to reference to video.

Do I need a GPU or any local setup to use MiniMax H3?

No. Generation runs on our infrastructure. You need a browser and an active plan, nothing else.

Which resolution should I generate at?

Draft at 480p while you are still deciding which idea survives, then rerun the winner at 768p for the version you actually publish.

Where to go next

If you are new to the model, start with what MiniMax H3 is and then walk through how to use it.