MiniMax H3 is the latest video model from MiniMax, the company behind the Hailuo AI app. It succeeds the Hailuo 02 generation and is available through MiniMax's API, the Hailuo app and partner platforms. MiniMax also published model weights on Hugging Face under its own community license.
MiniMax H3 specs
- Duration: 4 to 15 seconds.
- Frame rate: 24 fps.
- Audio: native 32 kHz stereo.
- Dialogue languages: stable support for 11 languages: Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian and Spanish.
- Aspect ratios: including 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16.
- Resolution: up to 2K through MiniMax's API (the 2K module is not part of the open release).
- Modes: text-to-video, first and/or last frame to video, and reference-to-video (up to 9 images, 3 videos and 3 audio clips; up to 12 files in total).
Is MiniMax H3 open source?
Partly. MiniMax released weights (a 33B-parameter transformer) under the "MiniMax H3 Community License Agreement," not a standard open-source license, with an application form for some regions. The model card notes that a hosted prompt-processing component, which it calls critical to quality, and the 2K module are API-only. Read the license before any commercial self-hosting.
What H3 is good at
- Talking scenes in many languages, thanks to multilingual dialogue and stereo audio.
- Controlled shots with start and end frames.
- Brand and character consistency with multiple image, video and audio references.
- Flexible formats, including cinematic 21:9 and vertical 9:16.
How to prompt MiniMax H3
- Specify language for dialogue: She says in Spanish, "¡Ya llegamos!"
- Assign each reference a role: "image 1 = character, image 2 = location, audio 1 = voice."
- Fit the action to 15 seconds or less; for longer stories, chain clips using the last frame as the next first frame.
- Describe the stereo soundscape: "traffic passes from left to right."
For general prompt structure see 30 text to video prompt ideas.
H3 vs other 2026 models
H3 tops out at 15 seconds, while Kling 4.0, Seedance 2.5 and Wan 3.0 reach 30 seconds. Its strengths are multilingual dialogue and published weights. See how it fits among the options in our Sora alternatives guide.
Check which MiniMax/Hailuo version is currently available in the TextVidAI generator.
FAQ
Is MiniMax H3 the same as Hailuo 3?
H3 is the model generation used in MiniMax's Hailuo AI products; many people call it Hailuo 3.
How long are MiniMax H3 videos?
Between 4 and 15 seconds per generation.
Does MiniMax H3 support sound?
Yes, native 32 kHz stereo audio with dialogue in 11 languages.
Sources
Last updated: October 9, 2026.