Kling 3 Motion Control transfers a recorded performance onto a character image. Instead of asking an AI video model to invent a dance, gesture, product presentation, or facial reaction from text, you provide the action as a reference video. The model then generates a new performance around the character pictured in your source image.
That distinction is the entire search job. Use ordinary Kling 3.0 when you want the model to invent motion from a prompt. Choose Kling 3 Motion Control from the current Cliprise model catalog when the movement already exists and must be reproduced with recognizable timing and intent.
On Cliprise, the current catalog price is 40 credits per generated second at 720p and 54 credits per generated second at 1080p. Treat the live model card as final truth before batching. A ten-second test therefore deserves more preparation than a disposable text-to-video draft.
Use Cliprise to run one short 720p proof first. If identity, limb motion, and framing survive that test, promote the same source package to 1080p.
Quick decision table
| Your job | Best starting route | Why |
|---|---|---|
| Recreate a dance or gesture | Kling 3 Motion Control | The reference supplies exact performance timing |
| Invent a cinematic shot from words | Kling 3.0 guide | Text-to-video has more creative freedom |
| Animate speech from a finished audio track | InfiniteTalk Audio2Video | Audio, not body motion, is the timing source |
| Preserve and restyle an existing clip | Wan 2.6 Video-to-Video | Existing video structure is the starting point |
| Make a simple camera move around a still | Image-to-video workflow | Full motion transfer is unnecessary |
What the model controls
Think of the inputs as separate departments. The motion reference supplies body movement, timing, expressions, and much of the orientation. The character image supplies the subject's visual identity, proportions, outfit, and starting composition. The text prompt should describe the setting, visual treatment, and details not already communicated by those assets.
This separation prevents a common error: writing a long prompt that contradicts the uploaded action. If the reference performer turns left while the prompt demands a fixed frontal pose, the model must choose between incompatible instructions. Motion transfer works best when the prompt supports the evidence already on screen.
Kling's official guidance says a source can be between 3 and 30 seconds and should be a continuous shot. It also recommends one performer, an unobstructed head and body, moderate speed, and no shot changes. Kling 3 adds stronger facial consistency and element binding compared with the 2.6 workflow, but it does not turn poor source material into reliable motion capture.
Build a reference package that survives generation
1. Choose a clean action clip
The ideal reference looks boring from an editor's perspective and excellent from a tracking perspective. Use a locked camera, one performer, an uncluttered silhouette, consistent lighting, and enough space around hands and feet. Avoid motion blur, jump cuts, whip pans, people crossing in front of the subject, or props that hide joints.
For a UGC presenter, record a single waist-up take with measured gestures. For dance, keep the entire body visible and reduce rapid travel across the frame. For a product demonstration, ensure the hands never disappear behind the object for long periods.
2. Match the character image to the action
A full-body dance reference needs a full-body character image. A close facial performance needs a head-and-shoulders source. This sounds obvious, but framing mismatch is one of the most expensive failure modes because the model must invent anatomy that the image never supplied.
Use a sharp image with a readable face, natural limbs, clean hands, and separation between the body and background. A neutral pose is usually more adaptable than an extreme pose. If the action turns sideways, an image with believable three-dimensional facial structure performs better than a flat, highly stylized profile.
3. Budget the test ladder
At the current Cliprise rate, resolution and duration directly affect the generation budget. Use this ladder:
- Crop the motion reference to the shortest useful performance.
- Generate one 720p test.
- Review the failure at full size, not only in a thumbnail.
- Fix the source package before changing five prompt variables.
- Generate 1080p only after the motion and identity are approved.
The broader credit optimization guide explains why cheap diagnostic passes beat repeated final-quality guesses.
A practical prompt structure
The action is already in the video, so keep the prompt compact:
[environment and time], [lighting], [camera treatment],
[wardrobe or material details that must remain],
[visual style], clean anatomy, consistent identity,
the subject remains fully visible throughout
Example for a fitness presenter:
Bright minimal fitness studio in morning daylight, stable medium-wide camera,
soft commercial lighting, black training outfit with no logos,
realistic skin texture, consistent identity and hands,
the performer remains centered and fully visible
Example for an animated brand character:
Playful grocery aisle with saturated practical lighting,
eye-level fixed camera, polished 3D character rendering,
preserve the red jacket and round glasses,
clean silhouette and consistent facial features during turns
Negative instructions can help when they are specific: no extra fingers, no duplicated limbs, no sudden camera cuts, no background people crossing the subject. A page of generic negatives usually adds less value than a cleaner reference.
Three production workflows
UGC and spokesperson motion
Record a performer delivering silent gestures to the intended pacing. Transfer that performance to the approved character, then add narration or dialogue in post. This separates body performance from voice timing and makes revisions manageable. If lip synchronization is the main requirement, route the approved portrait and audio through the InfiniteTalk guide instead.
Repeatable product demonstration
Create one clean reference in which hands enter, point, rotate the product, and leave. Keep the product interaction simple because contact points are difficult for generative video. Produce a 720p proof, inspect fingers and object geometry, then use the successful setup as a campaign template. The AI product demo workflow gives the surrounding edit structure.
Character performance library
For a recurring mascot or synthetic presenter, save approved reference actions by purpose: greeting, pointing left, pointing right, surprise reaction, short dance, and calm explanation. Standardize framing, duration, and safe areas. A small clean library is more useful than dozens of inconsistent clips.
Kling 3 versus Kling 2.6 Motion Control
Kling 2.6 remains useful when its lower current rate or established workflow matters. Kling 3 is the stronger starting point when facial identity must survive changing angles, occlusion, or more expressive acting. The current Cliprise catalog lists Kling 2.6 Motion Control at 22 to 36 credits per second and Kling 3 at 40 to 54 credits per second.
The decision is not simply newer equals better. Test 2.6 for broad body motion where the face stays readable and the budget is tight. Move to 3 when the face turns, expressions matter, or the output must withstand closer inspection. Read the Kling 2.6 Motion Control tutorial for the older route's controls.
Troubleshooting
The body movement is weak
Check whether the performer is large enough in frame, whether motion blur hides joints, and whether the reference contains continuous usable action. Increase clarity before adding adjectives such as dynamic or energetic.
The face changes during turns
Use a clearer source face and reduce the turn for the first test. If element binding is available in the selected workflow, supply matching front, side, and expression references. Do not expect one tiny frontal face to define every angle.
Hands or feet deform
Slow the reference, improve visibility, remove occluding props, and ensure the character image includes the relevant limbs. Crop the clip before the most complex contact if that moment is not essential.
The background moves unexpectedly
Use a simpler source image and ask for a fixed camera. Remember that matching the reference orientation can also carry camera behavior. Separate a complex camera move from complex body motion when possible.
The output is shorter than the reference
Very fast or discontinuous motion may leave fewer valid frames for extraction. Trim the source to a continuous moderate segment of at least three seconds. Credits spent on failed or shortened provider generations may not be recoverable, so test conservatively.
Quality-control checklist
- One primary character is visible throughout.
- Character and action framing match.
- The face remains identifiable at the first frame, midpoint, and final frame.
- Hands, feet, and object contacts are usable at playback speed and frame-by-frame.
- Clothing and key brand details do not change.
- Background motion does not distract from the performance.
- The output is checked in its final crop, including 9:16 safe areas.
- Synthetic people or performances are disclosed where policy or context requires it.
Kling 3 Motion Control is best understood as performance transfer, not a magic repair tool. A clean reference does most of the directing. The model maps that direction to a new character; Cliprise provides the practical place to compare the result with text-to-video, talking-video, and video-to-video routes without rebuilding the workflow.
Open the Cliprise model catalog, choose Kling 3 Motion Control, run the shortest 720p proof that contains the essential action, and only promote the shot after identity and anatomy pass review.