MiniMax
MiniMax H3
Capabilities
| Native audio | Yes |
| Multi-shot sequences | No |
| Image to video | Yes |
| Public API | Yes |
| Max resolution | 2K |
| Max duration | 15s |
Benchmark scores
| Overall | 4.81 / 5 |
| Prompt fidelity | 4.81 |
| Visual quality | 4.86 |
| Physics & motion | 4.38 |
| Subject consistency | 5.00 |
| Audio & sync | 5.00 |
| Average attempts to a usable clip | — |
| Average generation time | 448s |
| Average cost per run | — |
| Scored runs | 40 |
By prompt category
| Category | Score | Runs |
|---|---|---|
| Hands | 5.00 | 4 |
| Faces | 5.00 | 4 |
| On-screen text | 5.00 | 4 |
| Multi-shot | 5.00 | 4 |
| Lip-sync | 5.00 | 4 |
| Style | 5.00 | 4 |
| Image to video | 5.00 | 4 |
| Animals | 4.54 | 4 |
| Camera | 4.29 | 4 |
| Liquids | 4.25 | 4 |
The clips
Every score above links to the actual generation it came from. Nothing here is a mockup.
Hands avg 5/5
Chopping a pepper
Clean knife/pepper contact and grip, no artifacts.
Counting fingers
Exactly 5 digits throughout the fist-to-open-palm motion, verified frame by frame after the judge model claimed a sixth finger.
Fretting a guitar chord
Fingertips press strings against the fretboard cleanly, no clipping or fusion.
Tying shoelaces
Natural lace contact and knot-tying, all 5 fingers stable throughout.
Faces avg 5/5
Face in a crowd
Background pedestrians stay coherent, no merging or morphing.
Elderly portrait, slow blink
Wrinkle detail stays sharp through a realistic slow eye-opening movement.
Glasses reflection
Window reflection tracks with realistic parallax as the head turns.
Turn and smile
Identity stays coherent through the profile-to-camera turn and smile.
Liquids avg 4.25/5
Pouring coffee
Correct low-viscosity stream, turbulence and steam, no artifacts.
Honey drizzle
Correct high-viscosity behavior and light refraction.
Ice melting in whisky
FAILURE MODE: ice cube is pixel-identical between frames 2s apart, no melting or convection at all, verified directly.
Wave on rocks
Realistic spray atomization and gravity-driven drainage down the rocks.
On-screen text avg 5/5
Book chapter heading
CHAPTER ONE renders exactly and stays anchored through the push-in.
Neon sign OPEN 24H
Reads exactly OPEN 24H, stable through the flicker.
Bakery sign
FRESH BREAD DAILY renders exactly, no duplication or misspelling.
Printed t-shirt
SLOP renders exactly and tracks fabric movement naturally.
Multi-shot avg 5/5
Cafe, three shots
Coat, mug and table stay identical across all three shots.
Same bench, day and dusk
Bench geometry and framing stay locked across the midday-to-dusk change.
Watch, wide then macro
Dial pattern, indices and hands identical between wide and macro shots.
Alley, two angles
Jacket and alley environment match exactly from behind-shot to front-on turn.
Lip-sync avg 5/5
Bearded man, long line
Audio present, exact line, accurate phoneme lip-sync.
Direct address to camera
Audio present, exact line, accurate phoneme lip-sync.
Sung line
Audio present, sustained melody matches the sung phonemes.
Two-person exchange
Correct turn-taking and lip-sync; the model unprompted delivered the dialogue in Mandarin instead of English, with a faithful translation of both lines.
Camera avg 4.29/5
Crane to skyline
Genuine vertical rise revealing the skyline, occluded at frame 1.
Dolly in to a door
Clean perspective change, no warping on foreground walls/sconces.
Handheld follow
Organic handheld shake and footstep bobbing, not a smooth glide.
360 orbit
FAILURE MODE: camera stays essentially static (verified frame by frame, near-identical framing at start, middle and end), never attempts the 360 orbit.
Animals avg 4.54/5
Pigeon takeoff
Correct crouch, wing extension and push-off from the ledge, clean separation from stone.
Cat jumping to a counter
Real crouch, vertical leap, parabolic arc and balanced landing on the counter.
Dog shaking off water
Correct rotational shake propagating from head to spine, droplets follow real parabolic trajectories.
Galloping horse
Leg positions vary naturally frame to frame (an improvement over the repetitive cycle seen in other models), but hooves are heavily motion-blurred at the grass line, so genuine ground contact could not be confirmed either way.
Style avg 5/5
Anime walk cycle
Face and outfit stay identical, commits to clean cel-shaded anime style.
Claymation wave
Visible fingerprints and tool marks, avoids smooth CGI.
1970s film look
Genuine period grain, warm color timing and lens halation, not clean digital.
Rotating mug
Completes a full 360 rotation, handle stays attached and travels smoothly around.
Image to video avg 5/5
Animate a landscape
Terrain stays anchored, motion correctly restricted to clouds and grass.
Animate a vintage photo
Preserves vintage tonality and grain, no colorization or sharpening.
Animate a portrait
Face identity holds exactly through natural blink and head movement.
Product photo to spin
Invented geometry stays plausible and consistent through the full rotation.