Confirmed
- —Model name: MiniMax H3
- —Up to 2K output
- —4–15 second duration
- —Text, image, video, audio inputs
- —Up to 12 mixed reference files
Early review · Evidence-led
Our early MiniMax H3 review separates documented capability from demo observation—and leaves a blank where controlled testing should be.
Scorecard
A launch-day scorecard should reward documented product design—not pretend that curated demos equal a benchmark.
Documented support for text, image, video, and audio references.
Documented
Up to 2K and 15 seconds, with integer duration control.
Documented
Useful technical limits are public; pricing still needs live verification.
Partial
We have not run a controlled benchmark yet.
N/A
Fact check
Early verdict
On paper, H3's strongest idea is not one isolated quality metric. It is the breadth of context that can shape a video: text, boundary frames, visual references, video references, and audio references within one documented workflow.
That makes H3 especially interesting for iterative production, where continuity and reference control matter more than a single lucky prompt. But launch examples cannot establish failure rate, prompt sensitivity, or comparative value. Those require matched inputs, repeated generations, and recorded settings.
Our verdict: high potential, unusually flexible inputs, and no honest final score until independent tests exist.
Official H3 API integration
Validated generation briefs are sent through a server-only route to MiniMax-H3. Availability depends on provider capacity and credits; the site also labels its older Hailuo 2.3 fallback separately.
Build with clarity
Compose scene, camera, lighting, motion, and audio direction in one structured brief.
Open prompt builder