MiniMax H3
AI generationMiniMaxOmni-modal video with native stereo audio in one pass. Up to 15s, 768P or 2K. Wire nothing for text, an image to animate it (add a last frame to pin the end), or a driving clip plus up to 5 stills for a reference render.
- Runs
- On our servers
- Typical time
- 60-300s
- Inputs
- 9
- Outputs
- 1
Inputs
| Port | Type | ID |
|---|---|---|
| Prompt | text | prompt-in |
| First Frame / Ref 1optional | image | image-in |
| Last Frameoptional | image | image-2 |
| Videooptional | video | video-in |
| Audiooptional | audio | audio-in |
| Reference 2optional | image | image-3 |
| Reference 3optional | image | image-4 |
| Reference 4optional | image | image-5 |
| Reference 5optional | image | image-6 |
Outputs
| Port | Type | ID |
|---|---|---|
| Video | video | video-out |
Settings
| Setting | Type | Default | Options |
|---|---|---|---|
| Duration (seconds) | slider | 4 | 4 to 15 |
| Aspect RatioOutput shape | select | 16:9 | 16:9 Landscape, 9:16 Portrait, 1:1 Square, 4:3 Standard, 3:4 Portrait, 21:9 Ultrawide |
| ResolutionOutput resolution. 2K is regenerated from the 768P pass, so it carries more fine detail than an upscale. | select | 768P | 768P, 2K |
Prompting
Describe the scene, the motion, and the sound. H3 generates synchronized stereo audio natively, so dialogue, ambience and score can be directed in the same prompt.
Maximum prompt length: 5000 characters.
What this connects into (16)
Computed from the port compatibility rules, so it stays correct as nodes change.
Node ID minimax-h3. Open it in Studio with Tab, then type the name.