Kling 3.0 Motion Control does not invent movement — it takes the motion from a reference video you upload and re-performs it with your character.
Kling 3.0 Motion Control is the one model here that does not invent movement. Every other video model on this site starts from a description; this one starts from a performance you already have. Give it a character image and a reference video, and it re-performs that video's motion using your character. A text prompt is optional and does not steer the movement.
The limits follow from that design. If you do not already have a clip of the movement you want, this model has nothing to work from — use a generative model instead. You pay per second of the reference clip rather than an output length you choose, so a long reference is a large bill before you have seen anything. It is tuned for realistic human motion; heavily stylised 2D animation transfers less reliably. We have not benchmarked its speed.
Kuaishou's motion-transfer model maps movement from a reference video onto your character image — preserving face, expression, and identity with physics communities call mathematically dependable.
Upload a character image and a motion reference video; Kling transfers the movement — dance, gesture, action — onto your character while preserving facial identity and expression. The dance-transfer and character-swap workflows that fill social feeds run on exactly this.
Full-body tracking with accurate hand and finger articulation — reviewers specifically note individual knuckle definition where earlier versions collapsed hands into mush. Choreography and sign-level gesture work become viable.
Fighting, hugging, object interaction — high-difficulty contact physics render with a consistency users describe as mathematically predictable. Complex action that breaks other models is this model's home turf.
Generate at 720p or 1080p with per-second pricing shown upfront. Identity-stable, physics-true motion at broadcast-friendly resolution makes it a practical tool for creator content at scale.
Generated here on 15 August 2026, first take. The source photo and reference clip are themselves outputs from models on this site — no stock footage.
The subject performs the same head and shoulder movement as the reference, natural and unforced.
Transfer the performance timing from the reference clip while keeping the framing and identity of the source photo.
It is the model to reach for when you already have the movement and need a different person doing it.
Dance takes, gesture routines, a walk cycle you filmed on your phone — supply the clip and the character image, and the character performs it. This is the mechanism behind the dance-transfer and character-swap posts that fill short-form feeds.
Because the movement is fixed in the reference and only the character image changes, running the same clip against different images produces variants that actually match each other. A prompt-driven model gives you a fresh interpretation every time, which is the opposite of what this job needs.
Fighting, embracing, picking things up — the interactions that tend to fall apart when a model has to invent them are here inherited from real footage instead. Hand and finger articulation survives the transfer, which is what makes gesture work usable.
Keep Original Sound is on by default, so the music or dialogue that governed the timing of the original take stays attached to the output rather than needing to be re-synced in an editor.
If you do not already have a video of the movement, this model has nothing to transfer and you want Kling 3.0, Kling 3.0 Turbo or Seedance instead. It also has no aspect-ratio control, no 4K tier, and no scene generation — it changes who is performing, not what the scene is. Highly stylised 2D animation is the weakest case: it is tuned for realistic human motion.
The row that matters most is the first one: this is the only model here where the movement comes from footage instead of from a prompt.
| Kling 3.0 Motion Control | Kling 3.0 | Kling 3.0 Turbo | Seedance 2.5 | |
|---|---|---|---|---|
| What drives the motion | A reference video you supply | Your text prompt | Your text prompt | Your text prompt |
| Required inputs | A character image and a reference video — both mandatory | A prompt, optionally plus one first frame | A prompt, optionally plus one first frame | A prompt; images, videos and audio optional |
| Accepts a reference video | Yes — it is the point of the model | No | No | Yes, as a scene reference, billed on input plus output seconds |
| Resolutions | 720p · 1080p | 720p · 1080p · 4K | 720p · 1080p | 480p · 720p |
| Clip length | Set by the reference clip — up to 10s from image orientation, up to 30s from video orientation | 3–15s, any whole second | 3–15s, any whole second | 4–30s |
| Aspect ratio control | None — the output follows your inputs | 16:9 · 9:16 · 1:1 | 16:9 · 9:16 · 1:1 | Adaptive, plus six fixed ratios |
| Sound | Carries the reference video's own audio, on by default | Optional toggle, off by default | Always on, no toggle | Native synced audio, on by default |
| Credits per second, 720p | 83 | 45 silent · 60 with sound | 55 | 230 |
| Credits per second, 1080p | 110 | 60 silent · 85 with sound | 70 | Not offered — 720p is the ceiling |
| Credits per second, 4K | Not offered | 200, sound included at no extra cost | Not offered | Not offered |
| Credits for one clip, floor to ceiling | 249–3,300 | 135–3,000 | 165–1,050 | 440–6,900 |
| Speed we measured ourselves | Not measured | Not measured | 54s average per clip, fastest 29s (6 clips, July 2026) | 4s/480p ~2m 38s, 8s/720p ~5m 13s (7 clips, August 2026) |
Resolutions, clip lengths and every credit figure in this table are read from our own model registry on 20 August 2026 — the same per-second rate tables that price your generations, not vendor marketing pages. One difference worth reading twice: on the other three models the seconds you pay for are the seconds you asked for, while here they are the length of the reference clip you upload. Speed figures are our own timed runs on this platform, and every cell we have not measured says so rather than borrowing a number from somewhere else.
Compare Kling 3.0 Motion Control with other video models — same workspace, one click to switch.
| Model | kling-v3-motion-control, by Kuaishou Kling |
|---|---|
| What it does | Motion transfer — it re-performs the movement in a reference video using the character from your image. It does not generate motion from a description. |
| Required inputs | Two, both mandatory: one character image and one reference video. A text prompt is optional and does not drive the motion. |
| Character orientation | From Image — keeps your photo's framing, reference clip up to 10 seconds. From Video — follows the reference clip's orientation, up to 30 seconds. |
| Resolution | 720p or 1080p |
| Duration | Set by the reference video, not chosen separately. Minimum 3 seconds; the ceiling is 10 or 30 seconds depending on the orientation setting. |
| Aspect ratio | No selector — the output geometry follows your inputs |
| Sound | Keep Original Sound carries the reference video's own audio into the output, on by default |
| Credits | 83 per second at 720p, 110 per second at 1080p, counted against the reference video's length |
| Credit range per clip | 249 for a 3-second 720p transfer, up to 3,300 for a 30-second 1080p one |
| Measured speed | Not benchmarked — the two clips above returned in 144 and 166 seconds, which is two data points rather than a measurement |
Three steps from a blank prompt box to a finished clip.
Open the model picker and choose Kling 3.0 Motion Control. The panel shows its credit cost, resolution options, and duration limits before you spend anything.
Write the subject, the action, and how the camera behaves. Attach a first frame if you want the clip to start from a specific image.
Download the clip, extend it, or send it through further editing. Prompt, model, and settings stay attached in your history.
One account, the full image and video generation workflow.
Run Kling 3.0 Motion Control on the free tier's daily credits, or upgrade for premium models, faster generation, and 4K output.
Perfect for getting started
Billed annually · $240/year
Billed annually · $380/year
It transfers movement from a reference video onto your character image — keeping the character's face, expression, and identity while they perform the reference motion.
At minimum a character image and a motion reference video. You can choose whether the character's orientation follows the image or the video.
Per second of output at 720p or 1080p rates, displayed before generation. Failed generations refund automatically.
Dance and performance transfers, character swaps, AI-influencer content, and action sequences with complex contact physics.
Yes — there is an option to keep the reference video's sound in the output.
It is tuned for realistic motion; highly stylized 2D animation transfers are less consistent. For cinematic scene generation without motion reference, use Seedance or Veo.
One-click tools for everything around generation — polish, restyle, and share.
