MiniMax H3 overview
What Is MiniMax H3?
MiniMax H3 is a multimodal audio-video generation system developed by MiniMax. It can use text and, depending on the workflow, images, video, and audio as context. It generates 4–15 second audiovisual clips at 24 FPS with native stereo audio, starting from a 768p base result and supporting context-aware 2K regeneration.
Last verified: September 15, 2026
MiniMax H3 at a Glance
- Developer
- MiniMax
- Model type
- Multimodal audio-video generation
- Context
- Text, images, video, and audio
- Duration
- 4–15 seconds
- Resolution
- 768p base; up to 2K
- Frame rate
- 24 FPS
- Audio
- Native 32 kHz stereo
- Open checkpoints
- H3-Base FL2VA and Ref2VA
Which MiniMax H3 Workflow Should You Use?
Choose by the material you already have. FL2VA is for text and first/last-frame control; Ref2VA is for richer multimodal references.
| Starting point | Workflow | Guide |
|---|---|---|
| Text only | T2V | Open workflow |
| Image | I2V | Open workflow |
| First + last frame | FL2VA | Open workflow |
| Existing video | V2V | Open workflow |
| Multiple references | R2V / Ref2VA | Open workflow |
| Local UI | ComfyUI | Open workflow |
| Model weights | Hugging Face | Open workflow |
What Can MiniMax H3 Do?
Create from prompts or keyframes
Use text, a first frame, a last frame, or both.
Follow multimodal references
Guide identity, motion, camera behavior, or sound with image, video, and audio references.
Generate picture and sound together
H3 predicts video and audio jointly and outputs native stereo audio.
Produce 768p or 2K results
H3-Base outputs 768p; the complete system performs context-aware 2K regeneration.
What Languages Does MiniMax H3 Support?
MiniMax officially lists stable dialogue support for 11 languages: Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, and Spanish. Additional languages are supported to varying degrees.
This applies to spoken dialogue generation and does not guarantee equal pronunciation, voice quality, or visual lip synchronization for every language, speaker, prompt, or run. See the MiniMax H3 dialogue prompting guide.
| Stable dialogue languages | 11 |
|---|---|
| Korean dialogue | Supported |
| Japanese dialogue | Supported |
| English dialogue | Supported |
| Other languages | Supported to varying degrees |
| Perfect lip sync guaranteed? | No |
Source class: MiniMax primary documentation · Fact scope: MiniMax H3 dialogue generation · Last verified: September 15, 2026
Can MiniMax H3 Generate 4K Video?
Not as an officially documented native H3 output mode. MiniMax documents a 768p base result and output up to 2K through H3-Regenerate-2K. Native 4K generation is not currently listed as a standard MiniMax H3 output mode.
When a third-party platform exports an H3 video at 4K, distinguish native H3 generation, H3 2K generation, external upscaling, and post-processing. Official maximum currently documented: up to 2K.
Is MiniMax H3 the Same as Hailuo 3.0?
MiniMax H3 is the official model name. Hailuo AI is MiniMax's official consumer video product and provides access to MiniMax video models, including H3. “Hailuo 3.0” and “Hailuo H3” should be treated as informal wording unless current MiniMax primary documentation uses those names.
| MiniMax | AI company |
|---|---|
| MiniMax H3 | Video-generation model |
| Hailuo AI | Official MiniMax video product |
| MiniMax3.org | Independent third-party platform |
MiniMax H3 vs MiniMax M3
H3 is for audio-video generation. MiniMax M3 is a language, coding, and agent model. See the MiniMax M3 vs H3 comparison.
How Is MiniMax H3 Structured?
MiniMax describes H3 as a three-stage system: H3-Context-IR → H3-Base → H3-Regenerate-2K. H3-Context-IR interprets multimodal input, H3-Base generates the core 768p audiovisual result, and H3-Regenerate-2K uses that result together with the original context to produce the complete system's higher-resolution 2K output.
How to Access MiniMax H3
MiniMax H3 FAQ
What is MiniMax H3?
MiniMax H3 is MiniMax's multimodal audio-video generation system. It accepts text and, depending on the workflow, image, video, and audio context to create short video with native stereo audio.
What can H3 generate?
H3 can create video from text, animate a first or last frame, transform existing video, or follow image, video, and audio references.
Is H3 the same as M3?
No. H3 is MiniMax's audio-video generation system. M3 belongs to its language, coding, and agent model family.
Does H3 support audio?
Yes. H3 supports native synchronized 32 kHz stereo audio.
How long can videos be?
The official H3 release supports 4–15 second output. Available durations can vary by workflow and provider.
Does H3 support 2K?
Yes. H3-Base first generates a 768p result, then H3-Regenerate-2K uses that result and the original context to regenerate a 2K result.
Can I use images?
Yes. FL2VA supports an optional first frame, last frame, or both; Ref2VA supports image references in multimodal workflows.
Can I use video or audio references?
Yes. Ref2VA accepts text plus image, video, and audio references.
Where can I try H3?
Use H3 in a browser, through an API, or through supported local workflows. MiniMax also publishes H3-Base checkpoints on Hugging Face.
Can H3 run locally?
Yes. MiniMax publicly releases H3-Base FL2VA and Ref2VA checkpoints for supported local workflows. The downloadable H3-Base path is not the same as a fully local copy of MiniMax's complete hosted H3 system. See the MiniMax H3 Open-Source Guide for downloadable components and license restrictions.
What languages does MiniMax H3 support?
MiniMax officially lists stable dialogue support for 11 languages: Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, and Spanish. Additional languages are supported to varying degrees. This does not guarantee identical lip-sync or pronunciation quality.
Can MiniMax H3 generate 4K?
MiniMax officially documents 768p base output and up to 2K through H3-Regenerate-2K. Native 4K H3 output is not currently documented.
Is MiniMax H3 the same as Hailuo 3.0?
MiniMax H3 is the official model name; Hailuo AI is the official MiniMax video product. Treat Hailuo 3.0 as informal unless MiniMax documents it as official.