Kuaishou
Kling 3.0
Kling 3.0 supports multi-shot sequences on a shared audio timeline with native dialogue in five languages. Strong on high-motion scenes.
- Best category
- Image to video 5.00
- Weakest category
- Animals 3.75
- Clips scored
- 40
- Generation time
- 190s
What it is actually for
Kling 3.0 is where you go when the shot has to hold together across a cut. Its multi-shot support runs sequences on a shared audio timeline with native dialogue in five languages, and its subject consistency across camera angles is the strongest in the field.
That capability — the same character, recognisably the same, from behind and then front-on — is the frontier of 2026 and the thing that separates this generation of models from the last. Most models will quietly hand you a different person after the cut.
The trade-off
It is the strongest performer on high-motion scenes, but published pricing for it is a mess. In researching our Sora migration guide we found it quoted at $0.09–0.14 per second by one source and $0.029 by another, both published in 2026, both stated as fact. Check the provider directly and do not budget off a comparison table — ours included, until we have measured it.
Who should pick it
Character-driven and narrative work, short-form social where the same person appears across several shots, and anyone who needs frontier capability without frontier pricing. The Turbo variant trades some fidelity for speed and cost.
At a glance
Strengths
- The only model in our lineup with multi-shot support, generating continuous multi-angle sequences in one pass.
- 1 attempt on average per usable clip.
Watch-outs
- Lowest physics score among the top-performing models, 4.03/5.
- No native audio.
Multi-shot sequences where continuity across angles matters more than single-shot physics.
Capabilities
- Native audio
- No
- Multi-shot sequences
- Yes
- Image to video
- Yes
- Public API
- Yes
Benchmark scores
How we test →| Dimension | Score | |
|---|---|---|
| Overall | 4.47 / 5 | |
| Prompt fidelity | 4.31 | |
| Visual quality | 4.68 | |
| Physics & motion | 3.97 | |
| Subject consistency | 4.82 | |
| Audio & sync | — |
- Average attempts to a usable clip
- 1.00
- Average generation time
- 190s
- Scored runs
- 40
| Prompt category | Score | Runs | |
|---|---|---|---|
| Image to video | 5.00 | 4 | |
| Style | 4.92 | 4 | |
| Faces | 4.79 | 4 | |
| Camera | 4.79 | 4 | |
| On-screen text | 4.67 | 4 | |
| Multi-shot | 4.58 | 4 | |
| Animation | 4.50 | 4 | |
| Liquids | 4.46 | 4 | |
| Hands | 3.96 | 4 | |
| Lip-sync | 3.77 | 4 | |
| Animals | 3.75 | 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). Kling 3.0 benchmark results. Retrieved October 23, 2026, from https://genativity.com/model/kling-3-0<a href="https://genativity.com/model/kling-3-0">Kling 3.0 benchmark results</a> (Genativity)The clips
Every score above links to the actual generation it came from. Nothing here is a mockup.
Hands avg 3.96/5
Chopping a pepper
FAILURE MODE: knife grip shows finger warping; pepper slices duplicate, float and cling unnaturally to the blade.
Counting fingers
Five digits throughout. The index finger rises alone, then the other three unfold in quick succession and the thumb opens last.
Fretting a guitar chord
Correct anatomy, plausible fret/string contact, zero artifacts.
Tying shoelaces
Good finger anatomy, but the lace ties by morphing/fusing into a finished bow rather than real topological pulling.
Faces avg 4.79/5
Face in a crowd
Sharp subject, coherent background crowd.
Elderly portrait, slow blink
Natural blink, authentic wrinkle micro-texture.
Glasses reflection
Reflection looks somewhat anchored/painted onto the lens with limited parallax as the head turns, vs. tracking correctly on Sora.
Turn and smile
Identity fully coherent through the turn and smile.
Liquids avg 4.46/5
Pouring coffee
Accurate low-viscosity turbulence and foam; no gel artifacts (better physics than Sora on this one).
Honey drizzle
Correct high-viscosity laminar flow and pooling.
Ice melting in whisky
Correct refraction/meniscus but still little real melting; condensation appears abruptly rather than accumulating (milder failure than Sora's fully static render).
Wave on rocks
Convincing swell/break/runoff, slight blur in heavy foam.
On-screen text avg 4.67/5
Book chapter heading
FAILURE MODE: renders "CHARER ONE" instead of "CHAPTER ONE", a real character-substitution text failure. Stays stable/sharp but wrong.
Neon sign OPEN 24H
"OPEN 24H" exact, stable through atmospheric flicker.
Bakery sign
"FRESH BREAD DAILY" exact and stable.
Printed t-shirt
"SLOP" exact, moves naturally with fabric.
Multi-shot avg 4.58/5
Cafe, three shots
3 real cuts, woman/coat/cup/cafe identical throughout.
Same bench, day and dusk
FAILURE MODE: single static shot, no midday-to-dusk transition delivered at all, the prompt's core requirement (time change) is simply missing, though what is there is stable.
Watch, wide then macro
Delivered as one continuous push-in rather than two distinct shots (wide-then-macro) as prompted, though the watch itself stays perfectly consistent.
Alley, two angles
2 real cuts, jacket/alley identical throughout.
Lip-sync avg 3.77/5
Bearded man, long line
Same as above: mouths the line, silent clip.
Direct address to camera
Lips mouth the line correctly, but NO audio was generated at all. Fidelity docked for the missing spoken component; audio dimension left unscored (see model note), not failed.
Sung line
Mouth shapes and breath vapor present, but silent, no vocal track.
Two-person exchange
Correct turn-taking mouth movement between the two speakers, but silent, no audio track.
Camera avg 4.79/5
Crane to skyline
Textbook crane reveal with stable parallax, succeeds where Sora failed on this exact prompt.
Dolly in to a door
Clean dolly-in, fully rigid geometry.
Handheld follow
Good handheld follow; minor foot-morphing during fast strides.
360 orbit
Genuinely completes the 360 with stable object permanence, succeeds where Sora failed on this exact prompt.
Animals avg 3.75/5
Pigeon takeoff
Sound wing/foot anatomy, minor wingtip smearing on fast flaps.
Cat jumping to a counter
FAILURE MODE: the mid-air trajectory is simply cut, the cat teleports from launch straight onto the counter.
Dog shaking off water
Believable shake and spray; torso slightly too rigid instead of rippling head-to-tail.
Galloping horse
Airbrushed/artificial skin sheen; hooves float and slide with no real ground-contact force.
Style avg 4.92/5
Anime walk cycle
Consistent cel-shaded character, genuine 2D style.
Claymation wave
Convincing clay texture with visible stamped fingerprints.
1970s film look
Correct period grading and halation, but grain is milder than authentic 1970s film stock.
Rotating mug
Photorealistic studio look, rigid identity through rotation.
Image to video avg 5/5
Animate a landscape
Terrain static, clouds/grass move correctly.
Animate a vintage photo
Grain/tone preserved, no colourisation. NOTE: output rendered at 1656x1248 (~4:3), inheriting the source image aspect ratio instead of the requested 16:9, see model notes.
Animate a portrait
Faithful to source, natural blink and hair motion.
Product photo to spin
Full 180 degrees, consistent product identity. NOTE: output rendered at 1440x1440, inheriting the source image aspect ratio instead of the requested 16:9, see model notes.
Animation avg 4.5/5
3D feature film look
A convincing CG feature look with big glossy eyes, soft rim light and shallow focus. The robot follows the butterfly and raises its hand at 5 seconds, but the butterfly only hovers near the fingers and never lands.
Anime sword draw
Clean anime drawing, but for the first 4.5 seconds she stands still holding a bare blade, so there is no draw from the scabbard. She then raises it overhead and the clip cuts to a lunge against a new speed-line background at about 6 seconds.
Squash and stretch
The ball drops from a stone ledge rather than a table, flattens into a ring-shaped splat that holds for about 1.5 seconds, stretches tall and then spends two seconds with its back to the camera before the surprised face returns.
2D walk cycle
Clean side-view walk in place: feet plant, the body rises and falls on each step, arms swing against the legs, and the design holds for all 8 seconds.
Where to use it
The same model costs different amounts depending on the platform you reach it through.
| Platform | 10s costs | Max res | Watermark | Commercial use | |
|---|---|---|---|---|---|
 Higgsfield |
$0.99 | 4K | no | yes | Visit |
 Runway |
$2.5 | — | no | yes | Visit |
 Magnific |
— | 4K | no | yes | Visit |
 OpenArt |
— | 4K | no | yes | Visit |
 ImagineArt |
— | 1080p | — | yes | Visit |
 LTX Studio |
— | 4K | no | yes | Visit |
 Leonardo.Ai |
— | — | — | — | Visit |
 Krea |
— | — | — | yes | Visit |