Kling Motion Control: How It Works, Settings, and Prompts

Kling motion control takes two inputs, a reference video of someone moving and a character image, and produces a new clip where your character performs that exact movement. It copies body motion, facial expression, and pacing from the video while keeping the look of your image. On Oakgen it runs as Kling V3 Pro Motion Control and Kling V2.6 Pro Motion Control.

For clean results, use a single-subject reference clip of 3 to 30 seconds with no cuts and no camera movement, match the framing of your character image to the clip (full body to full body, half body to half body), leave Character Orientation on Match Video unless you need camera moves, and on V3 add a facial element when the face drifts. V3 Pro costs about 44 credits per second of reference video on Oakgen as of September 2026.

Kling motion control at a glance

Output length is the detail people miss most. There is no setting for it; the clip you upload decides it, so trim the reference first. The 30-second ceiling also only applies in Match Video mode. Switch to Match Image and anything over 10 seconds is rejected.

What motion control does

Motion control is a form of motion transfer AI: the model reads the pose and movement of the person in your reference clip and retargets it onto the character in your image. Kling describes it as control of a character's movements and expressions, extracted from an uploaded video.

People often mix it up with these:

• Image-to-video. You give a photo and a text prompt, and the model invents the motion. Fine for a slow smile or a camera push-in. Poor for choreography, because a sentence cannot describe 8 counts of a dance.
• Camera "motion control" prompting in Kling 3. Our Kling 3 prompting guide uses the phrase for directing camera speed, dolly moves, and subject direction in a text prompt. That is prompt vocabulary for regular text-to-video and image-to-video. The Motion Control model covered here is a separate mode that needs a reference video.

If you want to compare motion transfer against other reference-driven approaches, the Seedance 2.0 vs Wan 2.7 control guide covers how other model families handle reference motion.

Kling V3 Pro vs V2.6 Pro: the verified differences

Kling's guide says VIDEO 3.0 Motion Control "builds upon the Motion Control introduced in VIDEO 2.6" and focuses on facial consistency: stable features and smooth expressions in complex, multi-angle, and long-duration motion.

The concrete, checkable differences are smaller than the marketing suggests:

1. Facial element binding exists only on V3. You can attach one element made of a front-facing face image plus up to 3 extra reference angles. Kling says the element uses facial information only. It does not carry clothing, hairstyle, makeup, or props.
2. Element binding only works in Match Video mode. Kling, fal, and Oakgen's form all limit elements to that orientation.
3. V3 costs 50% more per second at the provider. $0.168 against $0.112 on fal. Kling's own API price list shows the same $0.168 per second for 1080p motion control.

Everything else (clip rules, orientation modes, the description field, the sound toggle, duration limits) is identical on both endpoints.

Which one to pick: use V3 Pro when the face has to hold up. Talking-to-camera spokesperson clips, close or medium framing, head turns, and anything where viewers know the person's face. V2.6 Pro is a reasonable choice for full-body dance shots where the face is small in frame, stylized characters, and rough drafts where you are still testing which reference clip works.

Every setting, explained

Character orientation

Most people leave it on the default without realizing it also decides who controls the camera and how long your reference clip can be.

• Match Video (default, up to 30 s). Per Kling, the character's movements, expressions, camera movement, and orientation all follow the reference clip. If the dancer in the clip turns to face left, your character turns left. Kling recommends it for complex motion.
• Match Image (up to 10 s). Movements and expressions still come from the reference, but the character keeps the orientation from your image, and camera movement is controlled through the prompt. Kling's showcase lists zoom in, zoom out, camera up, camera down, and a fixed position.

Rule of thumb: dance, sports, and any big body movement goes on Match Video. Use Match Image when your character is posed a specific way (a product shot, a seated presenter) and you want a camera move the reference clip does not have.

Description (optional prompt)

The reference video controls the motion; the prompt steers everything around it: wardrobe detail, background elements, lighting, props, and on Match Image, the camera. fal's schema notes that the characters, backgrounds, and other elements in the output are based on your image, so use the prompt for additions and adjustments rather than rewriting the scene.

Keep original sound

On by default. The output keeps the reference clip's audio track. Turn it off if the reference has music you do not have rights to publish, then add your own track in editing (more on that in the consent section).

Elements (V3 only)

Attach a front-facing face image, then optionally up to 3 more angles. Kling's advice is to upload faces that match the output you want:

Reference the element in your prompt as @Element1.

Reference clip checklist

Most failed generations trace back to the reference clip. Run every clip through this list before you upload. Each rule comes from Kling's guide or the model's input schema.

One practical note from Kling's guide: when the model cuts a result short because the motion was too fast, the Kling app does not refund those credits. Oakgen bills by the length of the reference clip you upload, so trim out the messy seconds yourself rather than hoping the model skips them.

Character image checklist

• Clear body proportions. Both fal and Oakgen's form say the character should avoid occlusion and take up more than 5% of the image area. In practice, bigger is better.
• Hands and limbs visible. If the reference waves and your character's hands are in their pockets, the model has to invent hands.
• Room to move. Kling says large motions need space around the character in the image. A tightly cropped portrait cannot do a wide arm sweep.
• Match the framing. A headshot driven by a full-body dance clip gives the model a scale problem it usually solves badly.
• Match the aspect ratio. Vertical clip, vertical image. Mixing a horizontal image with a 9:16 dance clip is an easy way to get warped bodies.

Need a character that does not exist yet? Generate a full-body shot on the AI image generator with a neutral pose, arms slightly away from the body, and plain background. If the character appears in more than one video, the AI character consistency guide covers how to keep them recognizable across images first.

How to use Kling motion control on Oakgen, step by step

1. Open Kling V3 Pro Motion Control (or pick Kling V2.6 Pro Motion Control from the video-to-video model list).
2. Motion Reference: upload your trimmed clip, or pick one of the 11 preset motion clips under the upload field if you just want to test.
3. Character: upload your character image.
4. Character Orientation: leave on Match Video for dance and full-body motion. Choose Match Image for a posed character plus a prompted camera move (10 s max).
5. Description: optional. Add scene, wardrobe, or lighting details. On V3 with an element, mention @Element1.
6. Advanced: decide on Keep Original Sound. On V3, add an element (front face first, then up to 3 angles) if identity matters.
7. Check the credit estimate, generate, and review the result frame by frame at the start, the fastest move, and the end.
8. If the output is soft, run it through the AI video upscaler as a separate step.

What Kling motion control costs on Oakgen

Oakgen passes through provider cost at 1 USD = 260 credits and charges V3 Pro by the length of the reference clip. As of September 2026:

For comparison, Kling V3 Pro image-to-video is priced at $0.336 per second in Oakgen's config, so a 5-second image-to-video clip is 437 credits. Motion control on the same model generation costs half as much per second, and it copies exact choreography instead of guessing.

Test with a 5-second slice of your reference before running the full 20 seconds. Most problems (wrong framing, limb glitches, face drift) show up early. V2.6 Pro lists lower at the provider; its exact price appears in the generator before you commit.

Motion control runs on Oakgen credits, which come with a paid plan.

Kling motion control prompt examples

Here the prompt only steers scene and style. Keep it short and concrete, and treat these as untested starting points to adapt.

1. Dance trend, clean background
A young woman in a cropped white hoodie and wide-leg jeans dancing in a bright minimalist studio, soft daylight from a large window, pale grey studio backdrop.

2. Add a background element (adapted from Kling's own example)
A corgi runs in and circles around the dancer's feet, cozy living room, warm evening lamp light.

3. Spokesperson with identity lock (V3, Match Video, element attached)
@Element1 presenting to camera in a modern kitchen, holding a matte black water bottle, natural window light, shallow depth of field.

4. Stylized character
A 3D animated fox character in a red scarf dancing on a snowy village street at night, string lights, gentle snowfall.

5. Product demo, posed character (Match Image)
A barista in a denim apron behind a coffee counter, slow camera zoom in toward the cup, morning light, steam rising.

6. Wardrobe swap while keeping motion
The same woman wearing a black satin evening dress and gold earrings, rooftop at dusk, city lights behind her.

A few rules that make prompts behave. Describe what you want to see, not the dance steps. Keep the character description consistent with the image (do not prompt "red hair" on a brunette). And for Match Image camera moves, name one move only.

Troubleshooting Kling motion control

Use cases that suit motion control

Dance trends. The obvious one, and the best fit, because a reference clip carries the exact timing of the routine. Film your own version of the trend on a tripod and use it as the reference, then apply it to an illustrated character, a mascot, or yourself in a different outfit. The AI dance video guide walks through the full photo-to-dance workflow, including music.

Product demos with a spokesperson. Film a team member doing the demo gestures (pick up the product, turn it, point at a feature), then map it onto a brand character. Use V3 with a facial element and Match Video for talking-head framing. Keep the product simple: motion control animates the character, and small product details such as labels can still drift, so check every frame.

Character animation and previz. Act out a fight beat or walk cycle on a phone and apply it to concept art. Rigging still does the final job. This just shows quickly whether a design reads in motion.

UGC-style ads. Record one creator-style take and apply it to several on-brand characters to test which look performs. A synthetic character should not claim a personal experience nobody had.

Multi-shot stories. Motion control generates one continuous shot. For a sequence, generate each shot separately with the same character image and element, then cut them together. The Kling 3 multi-shot character consistency workflow covers keeping a character stable across shots, and what Kling 3.0 brought to Oakgen covers the rest of the Kling 3 lineup.

Consent, likeness, and other people's dance videos

Faces and bodies. Animate a real person's photo only if it is you or you have their permission. Under Oakgen's terms, you need the rights and consents for any real likeness you upload, and using someone's likeness to impersonate or deceive without consent is not allowed. Sexualized content of real people is banned outright.

The reference video. Being public does not turn a TikTok into free stock footage. The recording belongs to whoever made it. On choreography, the US Copyright Office says choreographic works can be protected once fixed (for example, on video), while social dances, commonplace movements, and athletic movements may lack enough authorship to qualify. That is a line you should not try to judge case by case for commercial work.

The safe practice:

• Film your own reference, or use a friend who agrees to it.
• For commercial work, use licensed footage or get written permission from the creator.
• Credit the choreographer when you post a trend. That is a courtesy and grants you no license.
• Label synthetic people in ads and do not present them as real customers.
• Children's photos: get a parent's consent, and think twice before posting at all.

Limitations on Oakgen

• No resolution selector. Oakgen runs the Pro endpoints; there is no 720p versus 1080p choice in the form. Some other hosts expose one.
• No background-source toggle. Some hosts let you choose whether the scene comes from the reference video or the character image. Oakgen's form does not; the scene follows your image plus the prompt.
• Elements are V3 only. V2.6 Pro has no facial binding on Oakgen, because the provider endpoint does not offer it.
• One character per generation. Duets need two generations and compositing in an editor.
• Paid plans only. The free tier does not include motion control.

Sources

• Kling AI, Motion Control User Guide (VIDEO 3.0 and 2.6), accessed 2026-09-26
• Kling AI API pricing, Motion Control per-second rates, accessed 2026-09-26
• fal, Kling Video v3 Pro Motion Control model page and schema, accessed 2026-09-26
• fal, Kling Video v2.6 Pro Motion Control model page and schema, accessed 2026-09-26
• US Copyright Office, Circular 52: Copyright Registration of Choreography and Pantomime, accessed 2026-09-26
• Oakgen Terms of Service, accessed 2026-09-26
• Oakgen model configuration for Kling V3 Pro and V2.6 Pro Motion Control and Kling V3 Pro image-to-video (input limits, fields, and provider cost at 1 USD = 260 credits), checked 2026-09-26