AI Motion Control
NEWUpload one character image and one performance clip. The character re-enacts the performance.
Motion control replaces prompt-guessing with a real performance. You give the model a character image for the look and a video clip for the movement, and it renders your character doing what the clip does — same timing, same body mechanics, the clip's own soundtrack still attached. PixRoute runs this on Kling 3.0 Motion Control, with the bill computed from the reference clip's length before you start it.
Total is priced per second of your motion video — upload one to see it.
Nothing generated yet
Describe your idea on the left and hit Generate — output arrives here.
Two things are called “motion control”
The phrase arrives from two different places, and they want two different tools. This page is the second one. If you came here for the first, the link tells you where it lives on PixRoute.
Camera motion control
The film-set meaning: a robotic rig repeating an exact camera move, and the preset menus that imitate it — dolly in, orbit, crash zoom, bullet time. That is a property of the SHOT, not of a performer, and on PixRoute you get it by describing the move in the prompt on a text- or image-to-video model.
Go to Image to Video →Performance motion control
The meaning this page serves, and the one Kling 3.0 gave the name to: transferring a human performance from a video onto a character who was never filmed doing it. Dance, gesture, fight choreography, a walk cycle. Your image supplies identity and scene; the clip supplies the movement.
How motion transfer runs
Two uploads and a framing choice. There is no motion prompt to write — the video is the instruction.
Upload the character image
This defines who appears and what the shot looks like: face, outfit, proportions, background, style. Head and torso have to be fully visible — that is the one hard requirement the model enforces.
Upload the motion video
A 2–30 second clip of the performance you want copied. Its length sets both the output length and the price, so trim it before uploading rather than after.
Pick framing and run
Choose whether the output follows the video's framing or your image's, add a short prompt for style or mood if you want one, and start. Results land in My Creations.
Inputs that actually work
Most failed motion-control runs are input problems, and every rule below is one the pipeline enforces rather than advice.
Character image
- Head and torso fully in frame. Kling refuses an image where it cannot find a complete upper body — that refusal is the single most common failure on this tool.
- One clear subject, nothing crossing the face or chest.
- Short side of at least 300px, aspect ratio between 0.4 and 2.5 — the uploader checks both before the file is stored.
- Whatever is in the image comes along: the background and lighting of your output are the background and lighting of this file.
- Avoid recognisable real people. The model's own safety check refuses celebrity likeness on the input, and the run stops before it renders.
Motion video
- 2 to 30 seconds, up to 100MB. Longer clips are rejected at upload, not at render.
- One performer, with the full or upper body visible — head included — through the whole clip.
- Realistic footage transfers most predictably. The clip is read as body mechanics, so clean framing beats interesting camerawork.
- Its audio track is kept in the output. If the music matters to the piece, it is already handled; if it must not appear, strip it before uploading.
- Length is the bill. Ten seconds of reference costs five times two seconds, whatever the character image is.
The framing choice, and what it costs you
One setting, two behaviours, and a different length ceiling on each. It is the only creative decision the tool asks you to make.
| Match video motion | Match image framing | |
|---|---|---|
| Reference length ceiling | Up to 30 seconds | Up to 10 seconds |
| What the camera follows | The reference clip — its framing and camera movement carry over | Your character image — the composition you uploaded is held |
| Built for | Full-body choreography: dance, sports, fight beats, anything where the body leaves the frame's centre | Shots that must stay on your composition — a character speaking to camera, a product held in place |
| Output length | Follows the reference clip | Follows the reference clip |
AI Motion Control Pricing
Priced in credits, shared across every model on PixRoute.
Kling 3.0
Video to Video
| Credits / second | 30s reference |
|---|---|
| 40 | 1200 |
Motion control is the one mode on PixRoute billed by the INPUT: the rate above is charged per second of the reference video you upload, because that is what sets the output's length. Upload the clip and the exact figure appears on the run button before you spend anything. A failed run is refunded automatically — including the refusals in the table below.
When a run comes back empty
Every case below returns your credits. The point of listing them is that four of the five are fixable in the input, not in the settings.
- "No character detected"
- Kling's recogniser could not find a complete upper body in the character image. A head-and-shoulders crop, a figure turned away, or a subject occupying a sliver of a wide frame all trigger it. Use a half- or full-body image where head and torso are unobstructed.
- The input image was refused
- The safety check rejected the character image, most often for resembling a real, identifiable person. Nothing rendered and the credits came back; swap in an original or clearly fictional character.
- "Reference video must stay under 30s"
- The ceiling is per framing mode: 30 seconds when matching the video's motion, 10 when matching the image's framing. Either switch the framing pill or trim the clip — the check runs before you are charged.
- The upload itself failed
- Reference clips have to be 2–30 seconds and under 100MB, and the duration is measured server-side from the file, not taken from your player. A clip that reads as 30.4s will pass; one that reads as 31s will not.
- The price is missing until I upload
- That is on purpose. The bill follows the reference clip's length, so there is no honest number to show before the file exists. The moment it is measured, the run button carries the exact credit cost.
What motion control will not do
It transfers a performance onto a character. Four neighbouring jobs get confused with it, and each has its own page here.
Restyle an existing video
Turning footage you already have into anime, or repainting a scene while keeping its content, is a different operation. Motion control renders a new clip from your image.
What video to video means here →
Compose a scene from references
If you want a character, a product and a location combined into a shot that was never filmed, reference-driven generation is the tool, and it takes up to ten images.
Reference to Video →
Animate a photo on its own
No reference clip, just a still you want to move — with or without a closing keyframe. That is image to video, and every video model on the site runs it.
Image to Video →
Sharpen the result
Motion control returns the model's native resolution. To finish at 2K or 4K, hand the output straight to the upscaler — it takes a clip you already generated here.
AI Video Upscaler →
AI Motion Control FAQ
What does motion control mean in AI video?+
It means driving a generated clip with a real motion reference instead of a written description. You supply the movement as a video and the identity as an image, and the model renders the second performing the first. The term arrived from film, where a motion-control rig repeats an exact camera move; in AI video it has come to mean this performance transfer, which is what Kling 3.0 named its Motion Control feature after.
Do I still write a prompt?+
Only a short one, and not for the motion. The reference clip already specifies the movement; the prompt is where you note style or mood, or a change to the scene. Leaving it near-empty is a valid way to run this tool.
Can I use a clip I found on social media?+
Technically it will process; whether you may is a rights question we cannot answer for you. The performance, the music on the track and the performer's likeness can each be owned by someone. Filming the movement yourself, or using footage you have licensed, is the version of this workflow with no one else's rights in it.
Does it lip-sync my character to new speech?+
No. Motion control transfers the performance and keeps the reference clip's original audio — the mouth follows the movement in the source, not a new script. If you need a character speaking lines you wrote, that is a different kind of model and this is not it.
Why does the price change on every run?+
Because the reference clip's length decides both how long the output is and how much compute it takes. A 4-second reference costs a fraction of a 30-second one, and the exact number appears on the run button as soon as the clip is measured. There is no duration menu to choose from — the video is the duration.
Do the signup credits cover a motion-control run?+
Not a full one. New accounts get 50 credits, and motion control starts at roughly 80 for the shortest clip the uploader accepts, so this particular tool needs a top-up or a plan. The signup credits go a long way on the image tools and the cheaper video models — the credit table above is the honest way to plan a first session.
Have a clip whose movement you want?
Bring a character image and a performance under 30 seconds. The price is on the button before you spend anything.
Start a motion transferFree credits to start. No card required.