Product2026/07/31
Hailuo 03 (MiniMax-H3) Specs — Native 2K Video With Stereo Audio
Verified MiniMax-H3 / Hailuo 03 facts — native 2K, 4–15s clips, stereo audio, reference limits — and how Hailuo3 AI uses the model.
Hailuo 03 is the creator-facing name for MiniMax-H3 (also written Hailuo 3.0). MiniMax’s public API documents native 2K output, stereo audio, and 4–15 second integer durations. Hailuo3 AI is a focused web workspace around that model — not the MiniMax developer console.
Sources: MiniMax video generation guide, H3 API create, H3 launch blog.
Specs (official API)
| Item | MiniMax-H3 |
|---|---|
| Model ID | MiniMax-H3 |
| Output resolution | 2K (current API option) |
| Duration | 4–15 seconds, integers only |
| Audio | Native stereo with the video |
| Modes | Text-to-video · first/last-frame I2V · reference generation |
| Aspect ratios | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, or adaptive (mode rules apply) |
Reference limits
- Images ≤ 9 · videos ≤ 3 · audio ≤ 3 · mixed total ≤ 12 files
- Audio references must include an image or video (audio alone is rejected)
- Per-file size: image ≤ 30 MB · video ≤ 50 MB · audio ≤ 15 MB
- Prompt ≤ 7,000 characters
What’s new vs Hailuo 2.3
| Hailuo 2.3 (API line) | MiniMax-H3 / Hailuo 03 | |
|---|---|---|
| Resolution | Up to 1080p on documented 2.3 settings | Native 2K |
| Duration | e.g. 6s / 10s depending on resolution | 4–15s any integer |
| Audio | Prior line is more limited | Native stereo in generation |
| Inputs | Text + image paths | Text + first/last frame + image/video/audio references |
| Editing | Mostly regenerate | Instruction-based multimodal editing (model capability) |
Why it matters on Hailuo3 AI
- Native 2K masters without a separate upscaler step in your pipeline.
- Stereo audio so drafts leave the silent-prototype stage sooner.
- Text-to-video and first-frame image-to-video via one public HTTPS image URL.
- 5–15 second clips and prompts up to 4,000 characters in this workspace.
- A single-model workspace if you already know you want H3.
The official model capabilities above are not a list of enabled site features. Tail frames, multimodal reference packs, and instruction editing are not available here.
How to start on Hailuo3 AI
- Create an account, sign in, and choose a paid credit plan. Registration does not grant generation credits.
- Open the Hailuo 03 / MiniMax-H3 workspace.
- Choose text-to-video or first-frame image-to-video, write a prompt of at most 4,000 characters, and provide one public HTTPS image URL for I2V.
- Select 5–15 seconds and generate at 2K. Review the result in your account history and download it. Credits are reserved first and refunded on confirmed failure.
Ready to try it? Create an account →