Kling Motion Control: Transfer Motion and Swap Characters

Kling 3.0 Motion Control transfers the performance from a video onto a character image: you supply a reference image of the character you want on screen and a driving video of someone moving, and the model re-performs that motion with your character, keeping the choreography of the source clip. On VdoBloom it runs in the Animate Replace tool, costs 8 credits per second at 720p and 10.8 at 1080p, billed against a flat 10-second estimate — 80 or 108 credits — and takes driving videos from 3 to 30 seconds.

Updated October 2026.

What Kling 3.0 Motion Control is

It approaches character animation from the opposite direction to a normal image-to-video model. Instead of asking the AI to invent motion from a prompt, you give it two inputs: a reference image of the character, and a driving video whose human performance — gestures, timing, expressions, body movement — it transfers onto that character. The result keeps the choreography of the source clip with the identity of your image.

That is a different tool from text-to-video, and it is a different tool from lip sync. Where an avatar model moves a mouth to match audio, motion control moves a whole body to match a performance. You are not describing a movement; you are supplying one.

On VdoBloom there is only the Kling 3.0 version of Motion Control. There is no separate Kling 2.6 Motion Control variant, so if you see comparisons of the two elsewhere, they do not apply to this platform.

Inputs and limits

Input	Accepted
Reference image (your character)	JPG or PNG, at least 300px on the short side, aspect ratio between 2:5 and 5:2, up to 10MB
Driving video (the performance)	MP4 or MOV, 3 to 30 seconds, up to 100MB
Output resolution	720p or 1080p
Duration control	None exposed — output length follows the driving video
Aspect ratio control	None exposed — the reference image’s framing governs
Text prompt	Optional and short; the motion comes from the video, not the prompt

Two of those rows are the ones people trip over. There is no duration slider, so a 30-second driving video produces a 30-second-style performance rather than a trimmed one; and the 2:5 to 5:2 aspect window means an extreme panoramic reference will be rejected before the job starts. The 300px minimum is a low bar, but it exists — a small thumbnail will fail.

What it costs

Resolution	Rate	Billed estimate
720p (standard)	8 credits per second	80 credits
1080p (professional)	10.8 credits per second	108 credits

VdoBloom bills against a flat 10-second estimate for whichever resolution tier you pick, regardless of how long your driving video actually runs, and the exact number is shown before you generate. A failed generation is refunded automatically rather than silently spending the balance. For context on the same balance, VdoBloom’s pricing starts at 10 free credits on signup with no card, Lite is $15/month for 300 credits, and one-time packs start at $10.39 for 200 credits that never expire — paid output is watermark-free with commercial rights. At 80 credits a 720p clip, this is a hero-shot model rather than a sketch model: budget it the way you would budget a day of studio time, and rehearse on something cheaper first.

How to use it on VdoBloom

1. Open Animate Replace in the video-creation dashboard.
2. Pick the model and version. Choose Kling from the model dropdown and Motion Control as the version. The tab’s other options are Wan 2.2 Animate Replace (which works at 480p, 580p and 720p and is the default) and P-Video Replace, a budget character-swap model at 720p and 1080p.
3. Upload the driving video. MP4 or MOV, 3–30 seconds, under 100MB. This is where the performance comes from, so choose the clip for its movement, not its look — a phone video of someone dancing works better than a beautifully graded shot where nothing happens.
4. Upload the reference image. JPG or PNG, at least 300px on the short side, framing between 2:5 and 5:2, up to 10MB. There is also a library picker if you have characters saved.
5. Write the optional prompt if it helps. Keep it short. Describe what should stay consistent — wardrobe, hair, the setting — rather than trying to re-describe the motion you already supplied.
6. Pick 720p or 1080p and check the quoted cost. The price is calculated at 8 or 10.8 credits per second against the 10-second estimate and shown before you confirm.
7. Generate, review, and only then iterate. Because billing is per second of performance, the expensive habit is running the same clip repeatedly at 1080p while you fix the framing; fix it on the reference image first.

Example setups and prompts

With motion control, the setup is the creative decision and the prompt is a small consistency note. Six combinations that show the range:

1. Dance transfer. Reference: a full-body character illustration. Driving video: a short dance clip. Prompt: “keep the outfit and hair exactly as in the reference image, plain studio background.”
2. Product presenter. Reference: a person holding a product, cut out cleanly. Driving video: someone gesturing while talking, upper body only. Prompt: “keep the product and the label unchanged throughout.”
3. Sports move. Reference: a stylised athlete portrait. Driving video: a basketball shot in slow motion. Prompt: “same jersey colours across the whole move.”
4. Historical or fantasy figure. Reference: a costume character sheet in 2:3. Driving video: a modern clip of someone walking with purpose. Prompt: “keep the period costume and the consistent face from the sheet.”
5. Mascot performance. Reference: a brand mascot portrait. Driving video: a child waving. Prompt: “consistent fur colour, no added text or logos.”
6. Restaged archival motion. Reference: a character you own the rights to. Driving video: an old family clip of a gesture you want to preserve. Prompt: “clean background, keep the gesture and timing exactly.”

What the good ones share: the reference image is framed similarly to the driving video, the face is visible in both, and the prompt spends its words on consistency rather than on describing movement.

Getting a clean transfer: the mistakes that matter

• Mismatched framing. The most reported failure is face drift, and it happens when the reference image and the driving video frame the subject differently — a tight portrait reference against a wide full-body driving clip. Match the crop before you generate.
• More than one person in the driving video. The model has to decide whose performance to copy. Use a clip with a single clear subject.
• A low-resolution or heavily compressed reference. The 300px minimum is a floor, not a target. Use the largest clean image you have; a 10MB cap gives you plenty of room.
• Unclear rights. Motion control is the format most often used to put a real person’s performance onto a different character. Only use driving footage of someone who has agreed, and reference images you own or have licensed.
• Forgetting the content rating. Kling’s video models carry VdoBloom’s STRICT content label, which reflects provider-side filtering. Ambiguous reference material is the fastest route to a rejected job, and the refund policy does not make it a free experiment.

Which model to pick in the Animate Replace tab

The tab offers three tools, and the choice is mostly about fidelity versus volume. Wan 2.2 Animate Replace is the default and covers the most resolutions (480p, 580p, 720p), which makes it the cheapest place to test whether a character swap reads at all. P-Video Replace is the budget option at 720p and 1080p, running about 2.4 credits per second at 720p and 4.8 at 1080p — a 10-second clip is 24 credits at 720p or 48 at 1080p, well below Motion Control’s rates in the same tool. Kling 3.0 Motion Control is the maximum-fidelity path: it re-performs the source motion with your character and reaches 1080p, at 8 to 10.8 credits per second.

The practical sequence for a hero clip is to prove the swap cheaply, lock the reference image, then spend on Kling for the finished shot. If instead you want a different, easier job — swapping an object or an outfit in a clip while the rest of the shot survives — that is VdoBloom’s object swap route, covered in animate and replace explained.

Frequently asked questions

What inputs does Kling Motion Control need?

Two: a reference image of the character (JPG or PNG, at least 300px on the short side, aspect ratio from 2:5 to 5:2, up to 10MB) and a driving video whose motion will be transferred (MP4 or MOV, 3 to 30 seconds, up to 100MB). A short text prompt is optional.

How much does a clip cost?

8 credits per second at 720p and 10.8 credits per second at 1080p, billed against a flat 10-second estimate — 80 credits at 720p or 108 credits at 1080p for that estimate. The exact figure is shown before you generate and refunded automatically if the job fails.

Can I control the duration or aspect ratio?

No. Kling Motion Control exposes only a resolution choice, 720p or 1080p. Output length follows the driving video, and framing follows the reference image, which is why the reference crop matters so much.

Why does the face change during the clip?

Almost always because the reference image and the driving video frame the subject differently. Match the framing — both close-up or both full-body — and make sure only one person appears in the driving video so the model knows whose performance to copy.

Is there a cheaper alternative in the same tool?

Yes. P-Video Replace runs about 2.4 credits per second at 720p and 4.8 at 1080p, so a 10-second clip is 24 or 48 credits, and Wan 2.2 Animate Replace covers 480p, 580p and 720p if you would rather prove the swap before committing to a hero render. Reach for Kling Motion Control when the shot is the finished one.

Try it in Animate Replace, and see the Kling 3.0 Motion Control model page for the full spec. For the rest of the Kling family, Kling 3.0 video pricing covers the generation-side rates, and what to do when a Kling prompt is rejected applies to this model too.