Motion control takes two inputs that do different jobs: a picture of who, and a clip of what they do. The output is your character performing the reference clip's movement. It is the most reliable way to get a specific, complicated action out of a video model, because you are not describing the action — you are handing it over frame by frame.
Every other video lane asks you to describe an action in words, and words are a terrible format for movement. Motion control skips the description: it reads the movement out of a clip you supply and re-performs it with your character. A dance, a specific gesture, a walk with a particular gait — things no prompt reliably produces — become a matter of finding footage of them.
Two inputs, both required. The character image supplies identity, clothing and body; the reference clip supplies motion and timing. Neither is optional and neither substitutes for the other.
Which plan: Starter and up, paid in ✦ gold credits. Kling Motion Control is a hosted model billed per call by its provider, so green credits cannot pay for it at any length and a free account has no gold to spend.
| Property | Value |
|---|---|
| Character input | one image, face and body visible — required |
| Motion input | one reference video — required |
| Output length | 3 to 30 seconds, following the reference clip |
| Resolution | 720p |
| Wallet | ✦ gold, priced per second of output |
| Plan | Starter and up |
Thirty seconds is a long ceiling for this kind of model, and it is also the thing to watch on the bill: the lane is priced per second, so the reference clip's length is the invoice. Trim the reference to the movement you actually want before you submit it, not afterwards.
What it costs: Priced per second of output in gold. A five-second reference is a cheap test; a thirty-second one costs six times as much. The Generate button carries the exact quote and the wallet before you press it.
Open Motion — Opens the Motion section. Attach a character image and a reference clip; nothing runs until you press Generate.
There is a second motion lane, Wan Animate, and it runs on our own hardware. That normally means the green wallet pays — here it does not. Wan lanes are payable in gold only unless the account is Ultra, and Wan Animate additionally needs Creator. So on Starter and Creator it is a gold-billed lane like the hosted one, with no price advantage from running locally.
Watch out: Wan 2.2, Wan InfiniteTalk and Wan Animate are own-GPU models that the green wallet still cannot buy below Ultra. Do not plan a motion workflow around them expecting free credits to cover it.
| Kling Motion Control | Wan Animate | |
|---|---|---|
| Runs on | Hosted | Our own GPUs |
| Plan | Starter and up | Creator and up |
| Wallet | ✦ gold | ✦ gold below Ultra, ⚡ green on Ultra |
| Lengths | 3-30 s, following the reference | 4, 8 or 16 s |
| Resolution | 720p | 480p or 720p |
Motion control is usually blamed for identity problems that were decided by the input picture. The reference clip supplies movement; everything about who is moving comes from the one image, so it has to answer every question the render will ask of it.
The image composer is the obvious place to make that picture. A full-body figure on a plain background, generated at the ratio you want, is a cheap own-GPU render and it will save more gold than it costs.
A motion-control result is an ordinary video asset and every card action applies to it: regenerate with the original inputs intact, reuse those inputs on another model, boost it up the queue while it is still running, or hand it straight on to another section. Two hand-offs are worth knowing about — AI edit, which will re-shoot the clip from a written instruction if the setting or the wardrobe is wrong, and the lip-sync lane, if the character now needs to speak. Both are gold, both from Starter.
The order matters if you are chaining them. Get the motion right first, because a re-shoot through AI edit is capped at ten seconds and will not survive a thirty-second motion clip intact. If the finished piece needs to be longer than ten seconds and also needs an edit pass, do the edit on a short section and treat the rest as final.
Keep the inputs, too. The card holds the exact character image and reference clip that produced the render, so a second attempt at a different length or with a different prompt starts from what worked rather than from a folder on your desktop. That is worth more on this lane than most, because finding a reference clip that transfers cleanly is the slow part of the job.
Starter. Kling Motion Control is a hosted model billed in gold credits, and a free account is granted none. The own-GPU alternative, Wan Animate, needs Creator and is also gold-billed below Ultra.
Three to thirty seconds, following the length of the reference clip. Pricing is per second, so the reference's length is what you pay for.
720p. There is no resolution picker on this lane — the provider renders its native size.
Yes. Any video in your asset library can be the motion reference, including one you made in the composer. Trim it to the movement first.
Usually the reference and the character image disagree — a different body shape, a different framing, or more than one person in the reference. Match the two shots before blaming the model, and test on a short trim.