Narrative shots
Control how a scene begins, develops, and lands on its final composition.
MiniMax multimodal video
MiniMax H3 — Video Generation
Generate 4–15 second video at 768P or 2K from text, first and last frames, or image, video, and audio references.
Examples
See what MiniMax H3 can make
- Frame rate
- 24 FPS
- Duration
- 4–15 seconds
- Resolution
- 768P or 2K
- Audio
- Native stereo
- References
- Image, video, audio
From frames to finished motion
Define the shot before H3 animates it
Boundary frames establish the composition, the prompt directs the action, and H3 creates the motion between them. This makes the generation path easier to understand before you spend credits.
Create with MiniMax H3- Input1
Set the boundary frames
Upload a first frame, a last frame, or both to define where the shot begins and lands.
- Direction2
Describe the movement
Tell H3 what changes between the frames, including subject action, camera movement, timing, and sound.
“The car revs through the desert.”
- Output3
Review the generated shot
H3 builds a continuous clip while preserving the intended opening and ending composition.
What it is good for
MiniMax H3 — Video Generation
Generate 4–15 second video at 768P or 2K from text, first and last frames, or image, video, and audio references.
Create with MiniMax H3Character consistency
Use reference images to guide recurring subjects and visual identity.
Audiovisual concepts
Use video and audio references to shape motion, camera language, voice, and pace.
Key capabilities
Why creators pick MiniMax H3
01
Start and end on your frames
Animate an opening image, land on a final image, or use both to define the arc of a shot while H3 builds the motion between them.
“The car revs through the desert.”
Direct with multimodal references
Assign images to character or style, video to motion or camera language, and audio to voice or editing rhythm. H3 accepts up to 9 images, 3 videos, and 3 audio clips.
Draft motion, then finish in 2K
Explore timing at 768P, then render the selected direction at 2K. Choose any duration from 4 to 15 seconds and see the credit estimate before generation.
How to get a strong result
Tips for better results
Start with the subject, then add lighting, style, and anything that must stay the same. Upload a reference whenever you want a more controlled edit.
- 1
Assign each reference a role
Explain whether a reference controls character, motion, camera, style, sound, or timing.
- 2
Use both boundary frames carefully
Choose compatible first and last frames, then describe the transition between them.
- 3
Prototype at 768P
Validate motion and composition at lower cost before generating the final 2K version.
Related AI tools
Continue your workflow with other focused Video tools.
Series compare
Compare the MiniMax Hailuo series
Choose frontier multimodal control, high-fidelity generation, or a faster production variant.
| Feature | MiniMax H3Current | ||||||
|---|---|---|---|---|---|---|---|
| Best for | Multimodal cinematic direction | Fast, reference-led production | Maximum-speed iteration | MiniMax's high-fidelity video model for realistic human motion, cinematic effects, and expressive character performance. | A faster, lower-cost Hailuo 2.3 variant for image-to-video iteration and production volume. | MiniMax's efficient video model with text-to-video, image animation, and optional first-to-last-frame control. | A low-cost 512p Hailuo 02 variant for rapid concepts with optional first and last frame guidance. |
| Input | Text, image, video, or audio references | Text, frames, image, video, or audio references | Text or first and last frames | Text to video · Image to video | Image to video | Text to video · Image to video | Text to video · Image to video |
| Resolution | 768P or 2K | 480P, 768P, or 1080P | 480P, 768P, or 1080P | 768p / 1080p | 768p / 1080p | 512p / 768p / 1080p | 512p |
| Frame control | First and last frames | First and last frames | First and last frames | First frame | First frame | First frame / Last frame | First frame / Last frame |
| Speed | Flagship | Faster than real time | Fastest | Standard | Fast | Standard | Fast |
MiniMax H3 — Frequently Asked Questions
More models
Try another model
Explore other models, or browse the full catalog.
FLUX 3Video
Black Forest Labs' multimodal video model for synchronized-audio generation from text.
Veo 3.1Video
Highest-control Veo 3.1 option with subject references.
Veo 3.1 FastVideo
Faster high-quality generation with native audio.
Veo 3.1 LiteVideo
Cost-efficient option for higher-volume video workflows.