Hailuo 03 Is Here: MiniMax's Multimodal Video Model Lands on Apiframe

MiniMax released Hailuo 03 on July 31, 2026, and it is a bigger update than the version number suggests. It is also known as MiniMax H3. It is not just a faster or sharper Hailuo. It is a different kind of model: one system that understands text, images, video, and audio together, and can create or edit video from a single request. It is available on Apiframe now, behind the same API key and endpoint as the rest of the catalog.

What Is Hailuo 03?

Every earlier Hailuo release was a video generator and nothing more. You gave it a prompt, maybe a reference image, and it gave you back a clip. Hailuo 03 goes further. It can take mixed inputs at once (text, several images, video clips, and audio) and produce a result that respects all of them together. You no longer need separate tools for generating, editing, and referencing, and then stitch the results by hand.

In practice, three things matter most to developers:

Generation from mixed inputs: One request can use text, several reference images, video clips, and an audio track at the same time. Hailuo 03 blends them into one consistent shot.

Targeted editing: You can swap a character, replace a background, copy a voice onto a new speaker, or add a second person who moves in time with the first. Only the part you describe changes, and the rest of the shot stays the same.

Output you can ship: The model is tuned for commercial work, including readable on-screen text, product shots that stay on brand, app and interface motion, game cinematics, and short ads.

The Open-Weight Commitment

Most top video models are closed and available only through an API. MiniMax took a different path and published H3’s weights on August 3, 2026, on Hugging Face and ModelScope, under its own community license.

There are limits to how open this is. Only the base model was released. The two extra steps that produce the full 2K quality, a prompt preparation step and a 2K regeneration step, stay API-only. That means running the open weights yourself tops out at 768p, while the hosted API delivers full 2K. The community license also has conditions on who can use it and where, so read it before you plan to self-host.

For most teams, the hosted route is the simplest way to get full quality. If you use Hailuo 03 through Apiframe, you get 2K output without running any hardware yourself, and the open release is a useful backup if vendor risk is a concern for your product.

What's New, Capability by Capability

Omni Reference

Instead of a single reference image, Hailuo 03 accepts up to nine images, up to three video clips (2 to 15 seconds each, 15 seconds combined), and audio in one request. That is up to 12 reference files for images and video together, plus audio. Audio must be paired with at least one image or video, and cannot be sent alone. The model picks a matching aspect ratio from your mixed inputs, or you can set one yourself. This makes multi-shot brand campaigns, stories with the same character across shots, and style matching from a mood board practical in one call.

Native Audio, Every Time

Every Hailuo 03 video includes stereo audio by default: background sound, dialogue that matches lip movement, or music, all created along with the video. That removes an editing step most teams currently do outside the model.

First and Last Frame Control

You can give a start frame, an end frame, both, or neither. With no image at all, Hailuo 03 works as plain text-to-video.

Higher Resolution, Longer Clips

Clips can now be any whole number of seconds from 4 to 15, at 24 frames per second. Hailuo 2.3 only offered 6 or 10 seconds. Full output is 2K (1440p), and a cheaper 768p option is now available as well.

Instruction-Level Editing

Point the model at an existing video and describe the change: replace the cat with a dog, add one person walking in step on the left, or remove the green screen and add a fairy-tale forest. Hailuo 03 changes the part you named and leaves the motion, lighting, and pacing everywhere else alone.

Specs at a Glance

Where This Actually Gets Used

MiniMax built Hailuo 03 with commercial work in mind, and it shows in the kinds of prompts it handles well:

• Brand films and cinematic content: Trailers, commercial-style spots, and premium brand films built from a set of reference frames.
• Product and online store marketing: Clothing try-ons, product feature demos, and campaign videos that stay on brand across a whole shoot.
• Apps and interface design: Game menu mockups, mobile app motion, and landing page visuals with steady, readable text.
• Stories with the same cast: Vertical fashion videos, K-drama-style dialogue scenes, and short stories that keep a character’s face and clothes the same across shots.
• Editing existing footage: Swapping subjects, changing outfits, adding a character who moves with the scene, and replacing a green screen with a real setting, all on video you already have.

How to Use Hailuo 03 on Apiframe

If you're already using any model on Apiframe, this needs no new setup: same endpoint, same API key, same job pattern. Just point model at "hailuo-03":

curl -X POST https://api.apiframe.ai/v2/videos/generate \
  -H "X-API-Key: afk_your_api_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "a neon-lit night market street, slow dolly forward, steam rising from a food stall, ambient crowd noise and distant music",
    "model": "hailuo-03"
  }'

You get a job ID back immediately:

{
  "jobId": "b2c3d4e5-f6a7-8901-bcde-f23456789012",
  "status": "QUEUED"
}

Poll /v2/jobs/:id, or set a webhookUrl on the original request and let Apiframe notify your server when the clip is ready. Either way, a finished job looks like this:

{
  "id": "b2c3d4e5-f6a7-8901-bcde-f23456789012",
  "status": "COMPLETED",
  "model": "hailuo-03",
  "result": {
    "videoUrl": "https://cdn2.apiframe.ai/videos/b2c3d4e5-f6a7-8901-bcde-f23456789012.mp4"
  }
}

No MiniMax account, no separate SDK, no new billing relationship. Hailuo 03 sits behind the same one-API, many-models setup as Kling, Veo, Seedance, Runway, and every other video model already on Apiframe.

Pricing

Hailuo 03 is billed per second using Apiframe’s usual prepaid credits, where 1 credit equals $0.01. Full 2K output costs 22 credits per second. A 768p option starts at 10 credits per second for text-to-video. Reference video is billed at the same rate on top, so a 6-second clip guided by a 4-second reference video is billed as 10 seconds. The first five reference images are included, and each one after that adds 7 credits. Reference audio is free.

A plain 6-second clip at 2K therefore costs 132 credits, or $1.32. You only pay for successful generations, with no separate invoice and no rate card to negotiate. The model’s docs page has the full table. To see how this rate compares with the rest of the market, read our AI video API pricing guide, and use our guide to estimating monthly AI video costs if you are planning a budget.

Try It Today

Hailuo 03 is available now on Apiframe, alongside every other major video model, behind one endpoint and one API key. Get an API key and generate your first clip, or read the docs for full details on every setting.

Not sure which Hailuo model fits your project? Our guide to which Hailuo version to use compares the earlier versions side by side, and the Hailuo 2.3 API guide covers the previous generation in detail. You can also compare Hailuo API providers, see how it measures up in our FLUX 3 vs Hailuo 3 comparison, or browse our roundup of the best AI video generation APIs. Questions? Join our Discord.