MiniMax H3
MiniMax H3 (Hailuo 3.0) is the first omni-modal video model: mix text, image, video, and audio in one prompt, out comes 2K video with native sound. Best for reference-rich generation and open-weights builders.
The Spinoza Score
Reviewed by Kamran Arshad. See how we evaluate · Updated 2026-08-10.
- Type
- Video Model
- Category
- video generation
- Starts at
- Free
- Platforms
- —
- Launched
- Jul 2026
- Last reviewed
- 2026-08-10
A genuinely new kind of video model. Omni-modal input plus open weights make it the most flexible option in the category.
Hailuo built its name on physics-accurate motion and strong value, and H3 (Hailuo 3.0) pushes that into new territory. It's the first video model that takes text, images, video, and audio together in a single prompt and returns coherent 2K video with native stereo sound. Tag your references, up to nine images, three videos, and three audio clips, and it weaves them into the result. It also shipped as open weights, so builders can run and fine-tune it, outside a few excluded regions. Clips top out around 15 seconds, so it isn't the pick for long single takes, but for reference-rich, physics-solid video at a fair price, it's excellent.
- Reference-rich, multi-input generation
- Open-weights builders
- Physics-solid motion at a fair price
- Long single takes (caps around 15s)
- Users in regions excluded from local deployment
What it actually does
Omni-modal input
The first video model to take text, image, video, and audio together in a single prompt.
Native 2K + stereo audio
Renders at 2K with synchronized native sound generated alongside the video.
Tagged references
Combine up to nine images, three videos, and three audio clips, each tagged, in one generation.
Physics-accurate motion
Inherits Hailuo's reputation for believable body movement and natural facial micro-expressions.
Open weights
Shipped as open weights, so you can run and fine-tune it yourself (some regions excluded from local deployment).
Wide style range
Strong on anime, illustration, ink-wash, and game-CG looks, not only photoreal.
Composite of hands-on testing and editorial review — five dimensions, each rated 0–5 against the same published rubric for every tool in the category. How we score →
- Overall
- 4.5/5
- Motion & Realism
- 4.6/5
- Prompt Adherence
- 4.4/5
- Creative Control
- 4.4/5
- Audio & Sync
- 4.5/5
- Value
- 4.6/5
What it costs, by tier
Free credits in the Hailuo app; also shipped as open weights (local deployment excluded in some regions). Confirm current pricing.
Credits in the Hailuo app.
Self-host under its license (region limits apply).
Frequently Asked Questions
What is MiniMax H3?
MiniMax H3, also called Hailuo 3.0, is MiniMax's latest video model. It's the first omni-modal one, taking text, image, video, and audio as input and producing 2K video with native sound.
What does 'omni-modal' mean here?
You can mix inputs in a single prompt, text plus reference images, reference videos, and audio clips, and the model weaves them into one coherent result.
Is MiniMax H3 open source?
It shipped as open weights, so you can run and fine-tune it, though local deployment is excluded in some regions (US, EU, UK, South Korea).
MiniMax H3 or Seedance 2.5?
Seedance 2.5 leads on length and top-end control (30-second takes). H3 wins on omni-modal input, open weights, and value. Different strengths.
Ready to try MiniMax H3?
Omni-modal, 2K video with native sound.
Outbound link may be tracked. See affiliate disclosure.