# Kling AI Motion Control: Inputs, Credit Costs and Limits

> How Kling AI Motion Control works in November 2026: the 3 to 30 second reference rules, VIDEO 3.0 vs 2.6, credits per second and where it breaks.

_Source: https://professionalstoolkit.com/articles/kling-ai-motion-control — The Professional's Toolkit · updated November 2026_

---

Kling's Motion Control copies the movement in a short video onto one character in a still image. You upload a 3 to 30 second reference clip, or pick one from the built-in Motion Library, add a character picture and a scene prompt, and Kling returns a clip as long as your upload. It runs on two models. VIDEO 3.0 Motion Control, released in March 2026 with face binding, costs 9 credits a second in Standard mode and 12 in Professional; the older VIDEO 2.6 version costs 5 and 8 [1]. The verdict is simple. Use 3.0 when a face must stay recognizable, 2.6 for cheaper motion tests, and keep every reference to one continuous shot of one person.


**9-12 credits/s** — VIDEO 3.0 Motion Control cost per second of output, Standard to Professional mode.



- **3-30 s** — Motion reference length; output matches it
- **1 subject** — Copied per clip; largest person in frame
- **720p/1080p** — Motion Control output resolutions


## VIDEO 3.0 or 2.6: picking the right Motion Control

Motion Control is older than either model. It began as a beta on Kling 1.6 in June 2025, billed as extra credits on top of the generation. VIDEO 2.6 Motion Control launched in December 2025 with the 3 to 30 second reference window still in use today. VIDEO 3.0 Motion Control arrived in March 2026 and builds on it; Kling says its main gain is **facial consistency** across complex motion, multiple angles and longer sequences [2].

The mechanism is face binding. After the character image, you upload several photos or clips of the same face, ideally from different angles, and 3.0 binds them as an element. That element reads facial information only. Clothing, hairstyle, makeup and props still come from your character image, so a costume change belongs in the picture, not in the bound element. The first frame may show several people, yet only one element is supported. A stylized mascot with no likeness to protect can live with 2.6. A specific person held steady for 30 seconds needs 3.0.

One dated note on models. In November 2026 Kling's newest video model is Kling 4.0, released to all users on October 15, 2026, with a cheaper Kling 4.0 Flash. On the pages we read, Motion Control sits on VIDEO 3.0 and 2.6, not on 4.0, so choosing 4.0 in the model picker takes you into a different workflow. Our [Kling model comparison](/articles/kling-ai-models-compared) lays out what each model does.

## The three inputs and the checks each must pass

Kling builds every Motion Control clip from three inputs: a motion reference video, a character image and a text prompt for the scene. Most failed runs trace back to the first two. Check them against these rules before you spend credits.

| Input | Kling's requirement |
|---|---|
| Motion video length | 3 to 30 seconds; the output matches the upload |
| Motion video shot | One continuous take, no cuts or camera moves |
| Motion video size | Short edge at least 340px, long edge up to 3850px |
| Motion video subject | One dominant person |
| Character image | Short edge at least 300px; framing matches the video |
| Face binding (3.0 only) | Several photos or clips of one face |

The framing rule catches people most often. A full-body dance needs a full-body character image; Kling warns that a close-up face photo paired with a dance video fails, because the model has no arms or legs in the picture to move. Cuts are the other trap. A shot change or camera movement inside the reference can truncate the output, so trim a phone recording to a single take before uploading [3]. Kling also needs at least 3 seconds of usable motion to return anything.

The prompt controls everything the video does not dictate. Kling says text can still refine scene details while the motion stays consistent, so spend your words on setting, light and visual style rather than on the movement itself. The reference already supplies the choreography. Describing it again adds nothing and can pull the scene away from your image.


> 💡 **Match the framing first:** A full-body move needs a full-body character image. Kling warns that a close-up face photo paired with a dance video fails.


## Running a Motion Control job in the web app

The web app flow starts with the motion, not the character. Motion Control appears among the skills listed in the app header.

1. Open Motion Control and choose VIDEO 3.0 or VIDEO 2.6, then Standard or Professional mode.
2. Upload the reference action video from your files, or pick a clip from the Motion Library.
3. Upload the character image that should perform the motion.
4. On 3.0, add face-binding photos or clips if the likeness must hold.
5. Pick the orientation mode and write the scene prompt.
6. Read the credit cost on the Generate button, then confirm.

The orientation mode is the setting people skip. **Matches Video**, the default, makes your character follow the reference for movement, camera and body orientation, so a dancer turning toward the lens turns your character too. **Matches Image** keeps the orientation of your picture and hands camera movement to your prompt. Pick Matches Video for a faithful copy of a routine. Pick Matches Image when the still is a fixed mascot or product shot whose angle you want to keep, and describe the push-in or orbit yourself.

Output length follows the upload, so a 12-second reference returns a 12-second clip. Kling's feature page says Motion Control currently generates at 720p and 1080p, and the pages we read list no 4K option for it [4]. That ceiling matters if the rest of your edit runs on native 4K VIDEO 3.0 footage.

## What a Motion Control clip costs in credits

Billing is per second of output, rounded to the nearest whole second. Kling's own example: 3.4 seconds in 3.0 Standard costs 27 credits, while 3.6 seconds costs 36. By our arithmetic, a 10-second clip on VIDEO 3.0 costs 90 credits in Standard mode and 120 in Professional. On VIDEO 2.6 the rates fall to 5 credits a second in Standard and 8 in Professional. The exact figure appears on the Generate button before you confirm, so read it there.

Professional mode needs a paid plan, as do 1080p output, watermark removal and commercial use; free accounts generate at 720p with the Kling watermark. The entry Standard plan costs $8.80 a month or $79.20 a year and carries 660 credits a month. The Pro plan costs $32.56 a month per account and carries 3,000. Monthly credits expire after 31 days without rollover, and fees are not refunded once a plan is activated, both covered in our [Kling AI pricing guide](/articles/kling-ai-pricing).

One rule here differs from the rest of Kling. Failed generations normally return their credits. Motion Control carves out an exception: if complex or fast motion makes the output shorter than the upload, the credits consumed are not refunded. Fast choreography is exactly the input most likely to trigger that. Test the hardest eight seconds of a routine first, then commit to the full 30.


> ⚠ **Short outputs keep the charge:** If complex or fast motion makes a Motion Control output shorter than the upload, the credits consumed are not refunded, unlike ordinary failed generations.


## Where it breaks, and which Kling tool fits instead

Motion Control drives one subject. It transfers body movement and facial expressions, not speech; Kling's feature page says speech and lip sync need standard VIDEO 3.0 generation instead. A talking presenter with a gesture routine therefore takes two passes or a different tool. Group choreography is out too. With two dancers in the reference, the model copies only the one filling the largest part of the frame, and the partner disappears.

Some jobs that look like Motion Control belong elsewhere in Kling. For a dance that moves between two poses, Kling suggests uploading start and end frames to VIDEO 3.0 and describing the movement between them. To replace the dancer inside an existing clip, it points to VIDEO 3.0 Omni with a reference image of the new character. Kling 4.0's Omni Reference can borrow pacing, shot composition and camera movement from a reference video while swapping in your own characters and products, which suits ad recreations better than body-for-body copying. Kling describes generation as probabilistic, so budget for retries; our [Kling prompt guide](/articles/kling-ai-prompt-guide) covers how to cut down on them.


> 💡 **Pick the specialist tool:** Use VIDEO 3.0 start and end frames for a move between two poses, and VIDEO 3.0 Omni to replace a dancer inside an existing clip.


## Jobs it suits, and what to settle before publishing

Kling names a handful of uses: trending dances performed by a 3D avatar, animation rotoscoping, brand mascot ads and virtual acting. Its marketing guide adds sports actions and martial arts for TikTok and Reels, and says improved hand handling helps product demos. The same guide suggests a less obvious one, applying a single master camera move such as an orbit or push-in to many product images so a catalog of SKUs shares one look. Treat that as Kling's suggestion. It depends on clean product stills and Matches Image.

Length is the practical advantage for short-form work. A routine of up to 30 seconds fits one Motion Control generation, double the 15-second ceiling of a standard VIDEO 3.0 clip, so there is no seam to hide between two halves. That headroom is worth little if the reference has a cut at second nine. Shoot or trim the source as one unbroken take.

Three terms travel with every clip. Commercial use needs an active paid plan. Kling's terms ask you to disclose Kling-made output when you share it. And uploading a face or a performer's video counts as input, which Kling's terms let it use to train its models unless you revoke that permission by emailing its support address.


## FAQ

**Can Motion Control be used through Kling's API or an AI assistant?**

Yes. Kling's MCP server has a motion_control tool that takes a subject image plus a Motion Library motionId or a reference video, and the developer API lists Kling 3.0 motion control at 720p and 1080p with no 4K. API units are a separate currency from app credits.

**Does Motion Control work on animals or cartoon characters?**

Kling says it works on human, anime and 3D characters when the style stays consistent. A mismatch such as a human reference applied to an animal degrades the result.

**Can I use someone else's dance video as the motion reference?**

Kling leaves that check to you: its paid terms say users must clear portrait rights, copyright and trademarks before commercial use. Commercial use of any output also needs an active paid plan.


## The bottom line

Motion Control is the right Kling tool when you already have the movement on video and need a different character to perform it: one subject, one continuous take, matching framing. Pick VIDEO 3.0 at 9 or 12 credits a second when the face has to hold, and 2.6 at 5 or 8 when it does not. Expect 720p or 1080p, no speech, and no refund when fast motion shortens the result, so test hard routines in short sections before you commit a full 30 seconds.



## Sources

[1] Kling AI user guide: motion control user guide — https://kling.ai/quickstart/motion-control-user-guide (2026-11)
[2] Kling AI release note: Kling VIDEO 3.0 Motion Control launch — https://kling.ai/release-note/release-notes/4titicw2vg (2026-11)
[3] Kling AI blog: ai motion transfer video tutorial — https://kling.ai/blog/ai-motion-transfer-video-tutorial (2026-11)
[4] Kling AI feature page: ai motion control — https://kling.ai/feature/ai-motion-control (2026-11)
[5] Kling AI membership plans (US) — https://kling.ai/app/membership/membership-plan
[6] Kling AI Terms of Paid Service — https://kling.ai/docs/payment-policy
[7] Kling AI release notes — https://kling.ai/release-note/release-history
[8] Kling 4.0 user guide — https://kling.ai/quickstart/kling-4-model-user-guide