Genativity

MiniMax

MiniMax Hailuo 02

MiniMax Hailuo 02 ranks near the top of independent video leaderboards while sitting well below the frontier price point.

figures not yet verified by us

Watch the 44 clips Compare prices

Best category
Faces 5.00
Weakest category
Lip-sync 0.83
Clips scored
40
Generation time
300s

At a glance

Strengths

  • Subject consistency, 4.50/5, is its most competitive dimension, though every other model we have scored still beats it there too.

Watch-outs

  • Lowest overall score of any model we have benchmarked, 3.93/5.
  • No native audio (0.00/5), and the weakest physics score in the lineup, 3.66/5.
When to use it

Silent, low-motion clips only. MiniMax's own H3 outscores it on every dimension we measure, including audio, which Hailuo 02 does not generate at all.

Capabilities

Native audio
No
Multi-shot sequences
No
Image to video
Yes
Public API
Yes

Benchmark scores

How we test →
DimensionScore
Overall3.90 / 5
Prompt fidelity 3.84
Visual quality 4.14
Physics & motion 3.59
Subject consistency 4.50
Audio & sync 0.00
Average generation time
300s
Scored runs
40
Prompt categoryScoreRuns
Faces 5.00 4
Multi-shot 5.00 4
Image to video 5.00 4
Liquids 4.17 4
Animation 4.13 4
Style 3.96 4
Hands 3.88 4
On-screen text 3.88 4
Animals 3.71 4
Camera 3.63 4
Lip-sync 0.83 4

Scores out of 5, judged blind. Bars start at 3.0 so the gaps are visible.

Cite this

Free to cite and reuse with a link back. Data is updated as new runs and prices come in, so include the date.

Genativity. (2026). MiniMax Hailuo 02 benchmark results. Retrieved October 23, 2026, from https://genativity.com/model/hailuo-02
<a href="https://genativity.com/model/hailuo-02">MiniMax Hailuo 02 benchmark results</a> (Genativity)

The clips

Every score above links to the actual generation it came from. Nothing here is a mockup.

Hands avg 3.88/5

Chopping a pepper

Realistic grip and knife motion, but a slice clings to the blade and slides along it instead of falling.

Counting fingers

A normal five-finger hand, but the fingers open together rather than one at a time.

Fretting a guitar chord

The fretting hand is fine, but the right-hand fingers never touch the strings and the shot is nearly static.

Tying shoelaces

Hands stay on-screen and tie correctly throughout; the only real artifact is one lace segment held stiff and straight rather than draping naturally near the end.

Faces avg 5/5

Face in a crowd

Background crowd shows correct occlusion and depth blur, no merging or morphing.

Elderly portrait, slow blink

Wrinkle texture persists through two natural slow blinks.

Glasses reflection

Lens reflections shift plausibly with head turn, identity stays stable.

Turn and smile

Identity, earring and features stay stable through the profile-to-camera turn.

Liquids avg 4.17/5

Pouring coffee

Accurate stream breakup, refraction and steam dispersal, no artifacts.

Honey drizzle

Correct laminar stream, necking and thinning as the dipper lifts.

Ice melting in whisky

FAILURE MODE: whisky, ice cube and condensation droplets all stay completely static for the full clip, no melting or convection at all.

Wave on rocks

Natural crest, shatter and foam wash, no looping or static textures.

On-screen text avg 3.88/5

Book chapter heading

FAILURE MODE: hallucinated gibberish subtitle text (COLF OOL dT, AI II JDBAXSOI...) alongside the correct CHAPTER ONE header.

Neon sign OPEN 24H

Reads exactly OPEN / 24H, stable through the flicker.

Bakery sign

FAILURE MODE: header reads BRRAANTOP BREAD and side signs misspell DAILY as DALY, instead of FRESH BREAD DAILY.

Printed t-shirt

SLOP renders exactly and stays anchored to the fabric as the person walks forward.

Multi-shot avg 5/5

Cafe, three shots

Coat, table and cafe interior match seamlessly across all three shots.

Same bench, day and dusk

Bench geometry and framing stay locked, only ambient lighting changes.

Watch, wide then macro

Case, bezel, dial layout and strap identical between wide and macro shots.

Alley, two angles

Jacket and alley environment stay continuous from behind-shot to front-on turn.

Lip-sync avg 0.83/5

Bearded man, long line

FAILURE MODE: no audio track; mouth stays mostly closed with minimal jaw movement, doesn't attempt to articulate the line.

Direct address to camera

FAILURE MODE: no audio track at all; mouth does generic open/round/close flapping unrelated to the intended line.

Sung line

FAILURE MODE: no audio track; mouth shows generic singing-style morphs unrelated to the actual sung phrase.

Two-person exchange

FAILURE MODE: no audio track; both speakers show generic alternating mouth flapping, no phoneme match, turn-taking only loosely implied.

Camera avg 3.63/5

Crane to skyline

FAILURE MODE: camera stays static instead of rising; skyline is stable but shown fully from frame one rather than revealed.

Dolly in to a door

Clean perspective change, door scales up without warping.

Handheld follow

Authentic handheld wobble, child stays anatomically coherent while running.

360 orbit

FAILURE MODE: camera stays essentially static, never attempts the 360 orbit around the subject.

Animals avg 3.71/5

Pigeon takeoff

Wingbeat, neck and push-off all anatomically correct through the full takeoff sequence, verified frame by frame.

Cat jumping to a counter

FAILURE MODE: cuts away with a whip-pan during the jump itself, never shows the ballistic leap the prompt asked for.

Dog shaking off water

Realistic inertial body wave and droplet ballistics, no artifacts.

Galloping horse

Hooves do touch the ground at points in the stride, but the body shows no vertical bob across the gait cycle, giving the gallop a mechanical, looping feel rather than true stride physics.

Style avg 3.96/5

Anime walk cycle

Face and outfit stay identical, commits fully to flat cel-shaded anime style.

Claymation wave

Genuine hand-molded clay texture with tool marks and stop-motion feel, not smooth CGI.

1970s film look

Warm color grading and lens halation land, but the image stays sharp/clean digital rather than true grainy period film.

Rotating mug

FAILURE MODE: mug stays largely stationary instead of rotating; handle subtly morphs and widens rather than traveling.

Image to video avg 5/5

Animate a landscape

Terrain stays static, motion correctly restricted to clouds and swaying vegetation.

Animate a vintage photo

Stays black-and-white and grainy, no colorization or sharpening.

Animate a portrait

Face identity holds exactly through head turn and natural blinks.

Product photo to spin

Invented back geometry of the mug stays plausible and proportion-consistent through the rotation.

Animation avg 4.13/5

3D feature film look

A glassy blue robot with screen eyes that squint with emotion. The butterfly lands on the raised hand early, at about 1.5 seconds, and the clip is 6 seconds long.

Anime sword draw

A soft, painterly anime look with an overhead draw and a cut to a side-view lunge with speed lines. During the lunge the blade ripples like a ribbon instead of staying rigid. The clip is 6 seconds long.

Squash and stretch

Big, readable squash and stretch, from a flat pancake to a tall cone, but the ball lands on the table and never hops off its edge. The clip is 6 seconds long.

2D walk cycle

Starts as a walk but the fox travels across the frame and at about 4.8 seconds turns round to face the camera and then walks the other way, so the side-view cycle breaks. The clip is 6 seconds long.

Where to use it

The same model costs different amounts depending on the platform you reach it through.

Platform10s costsMax resWatermarkCommercial use
Higgsfield $0.7 1080p no yes Visit
Magnific — 1080p no yes Visit
ImagineArt — 1080p — yes Visit
Krea — — — yes Visit
OpenArt — 1080p no yes Visit
Lumeta — 1080p no yes Visit