Kling 3.0 Pro Motion Control on fal: How to Copy a Dance Onto Your Own Character (October 2026)
Motion Control reads the movement out of a reference clip and re-performs it with your character, at $0.168 per second. The orientation setting that decides identity, the full-body requirement behind most 422 errors, and where Scail 2 now competes.
Kling 3.0 Pro Motion Control is the endpoint behind most AI dance videos you have seen with a real person's face on them. It takes a reference clip, extracts the motion path, and re-performs it with a character you supply as an image. It lists on fal at $0.168 per output second (the Standard tier is $0.126), it is fal's own pick for motion transfer, and it has been the default for dance in our pipeline since April. This guide covers the two settings that decide the result, the input rule that causes most errors, and the first serious challenger.
What it costs
| Output length | Pro ($0.168) | Standard ($0.126) |
|---|---|---|
| 5 seconds | $0.84 | $0.63 |
| 8 seconds | $1.34 | $1.01 |
| 10 seconds | $1.68 | $1.26 |
| 30 seconds | $5.04 | $3.78 |
Pro over Standard is worth the 33 percent whenever the face is on screen. In our side-by-side runs Pro held identity visibly better, and an extra 40 cents on a ten-second clip is cheaper than a rerun.
The setting that decides everything: character_orientation
The endpoint requires character_orientation, with two values:
video: the reference clip drives the framing and the camera, and the model honours theelementsentry that binds your face to the character. The reference can be up to 30 seconds.image: the framing follows your still. In this mode the face binding is effectively a no-op: the request succeeds, you are billed, and the face drifts.
If you try Motion Control and conclude "Kling drifts faces", check this value first. It is the most common cause by a wide margin, and the one we most often see in other people's setups.
The input rule that triggers errors
Both inputs need to show a full upper body. In early October we saw 422 errors on stills cropped at the shoulders and on reference clips that cut to a close-up. The practical checklist:
- The character photo: full body if the dance uses legs, feet visible, standing, neutral pose, plain background. Anything the photo does not show, the model invents, and the invention is where viewers' eyes go.
- The reference clip: one dancer, upright and repeating movement, no jump cuts into close-ups. Floor work and rapid direction changes transfer badly.
- Length: five to eight seconds of the recognisable section. Drift compounds with duration multiplied by motion amplitude, and dance maximises the second term.
What it does well, and what it does not
It re-performs motion with cloth and hair dynamics rather than pasting movement on a rig, and it keeps the reference's timing. It does not read the audio, so re-attach the trending track yourself, and it does not license anything: the choreography, the soundtrack and the likeness are three separate rights questions, covered in can you remake someone else's Reel.
It is also not a face swap. The whole body is regenerated, so wardrobe and proportions come from your image and the prompt rather than from the source dancer. That is why it works for the AI baby dance format, where mismatched proportions are the joke.
The challenger: Scail 2
Scail 2 lists $0.10 per second, 40 percent below Kling Pro. In a blind pair we ran in August, we preferred its output on that clip. Its limits explain why Kling remains the default: output is 512p or 704p at 16 frames per second, long driving clips are truncated, and it needs an upscale pass to reach 1080p. We cover those in the Scail 2 guide. Other options at the low end are Wan Motion and One-to-All Animation, both listed around $0.06.
Choosing between Motion Control and replacing the person
If the Reel has a cut every second or two, or several people, motion transfer is the wrong tool, because it follows one body through one continuous motion. A video-edit model (Kling O3 edit, Recast, HappyHorse video-edit) keeps the original cuts instead. Our dance-versus-Recast comparison lays out when each applies.
Common questions
Why does my face drift with Motion Control?
Check character_orientation first. With the value image the framing follows your still and the face binding is a no-op; with video the reference clip drives framing and the face element is honoured.
How long can the reference clip be?
Up to 30 seconds in the video orientation. We still recommend five to eight seconds of the recognisable section, because drift compounds with length.
Why do I get 422 errors?
In early October we saw them when the character photo or the reference clip did not show a full upper body. Use a full-body still and a reference without close-up cuts.
Does it add the music?
No. Re-attach the trending sound yourself and check the licence, because the movement is generated and the soundtrack is not.
How to do this without code: Motion Control is the model behind the Dance copy tool, so you need neither an API key nor the orientation setting. In AutoHustle this is the Dance copy tool: upload the reference clip and one full-body photo (3 credits per second with a 24-credit minimum, quoted before you press Generate). See How to make an AI dance video and AI dance video generator.
Related reading: Why Your AI Dance Video Looks Wrong (Five Causes, In Order) · Kling 3.0 Pro Image-to-Video on fal: Price, Elements and Limits (October 2026)
For developers: the fal endpoints
| Model | fal endpoint |
|---|---|
| Kling 3.0 Pro Motion Control | fal-ai/kling-video/v3/pro/motion-control |
| Kling 3.0 Standard Motion Control | fal-ai/kling-video/v3/standard/motion-control |
| Scail 2 | fal-ai/scail-2 |
Required: a reference video, a character image, and character_orientation (image or video). The elements entry (one facial element) is honoured only with video. Prices read from fal on 18 October 2026.