AI Motion Control
Give it one reference video and your characters. Motion control takes the movement, camera angle and pacing from the clip — the faces, hair and bodies come from your images.
Motion Control
MP4 or MOV. Reference video length counts towards the price — 8 credits per second, so a shorter clip costs less.
Reference images in your prompt
Upload a reference video to see the price.
Create Anything You Can Imagine
Four inputs, one clip. Motion control reads how things move from the reference video; your images decide who is moving.
1. Reference Video
2. Character Images
image 1
image 23. Write Your Prompt
Two-person scene. The woman in image 1 and the man in image 2 perform the action of video 1. Use video 1 as the motion template: reproduce its choreography, body positions, points of contact, rhythm and pacing, and its camera framing and angle — but the two performers are replaced by the woman from image 1 and the man from image 2, and their faces, hair, skin tone and body proportions stay unchanged. Opening arrangement: warm lamplight in a bedroom, he holds her close against him on the bed, his penis inside her vagina, one hand spread across her lower belly and thigh, her head tipped back with a soft smile. Continuous motion in sync with the tempo of video 1, her breathing and quiet moans timed to it. Amateur phone-camera look, warm practical lighting, shallow depth of field, handheld intimate point of view.
4. Get Your Video
Upload a clip whose motion you want, add the characters who should perform it, and describe the scene. AI motion control keeps the choreography, camera and pacing of your reference while putting your own characters in it.
How AI Motion Control Works
Motion control separates two things that usually travel together in a video: how a body moves, and whose body it is. The reference clip you upload supplies the first — its choreography, its contact points, its rhythm, its camera angle. Your reference images supply the second — the faces, hair, skin tone and proportions the model must keep.
The model then regenerates the scene rather than editing your clip frame by frame. That is the practical difference between motion control and a video editor: nothing from the original footage survives into the output except the movement itself, so the people in the result are yours and the frames are new. Because motion control regenerates the scene, the output does not inherit your reference clip's quality: a phone-shot reference and a studio one render at the same fixed resolution. What the clip does carry across is its framing, so a vertical reference yields a vertical result.
Motion transfer is the technical name for the same pipeline — motion transfer describes the mechanism, while motion control describes what you get to steer with it. If you came here searching for either term, this is the tool you were looking for.
The video sets the movement
One reference clip, up to 15 seconds. Motion control reproduces its pacing, body positions and camera framing — you do not have to describe any of that in the prompt.
The images set the identity
One to four character images. Name each one in the prompt — the woman in image 1, the man in image 2 — and the model binds each face to the right person in the scene.
The price is visible first
Motion control bills for the reference clip as well as the output, so the cost is measured the moment your upload finishes, before a single credit is spent.
Reference Video Requirements
These come from the model, not from us, and every one of them is checked on our side before you are charged. A file that fails any line is rejected at upload rather than after a generation has already been paid for.
- Container format
- MP4 or MOV
- Reference clip length
- 1–15 seconds
- Reference clip + output
- ≤ 30 seconds
- File size
- ≤ 50 MB
- Character images
- 1–4, ≤ 10 MB each
- Output length
- 5 / 10 / 15 seconds
Motion Control vs Face Swap
Three tools on this site put a person you supply into a video, and they are not interchangeable. Motion control is the only one that takes both a clip and a character set, which is also why it is the only one whose price depends on what you upload.
| Motion Control | Reference to Video | Video Face Swap | |
|---|---|---|---|
| What you upload | 1 video + 1–4 images | 1–4 images | 1 video + 1 face |
| What is replaced | The whole person | Nothing — built from scratch | The face only |
| Where the motion comes from | Your reference clip | Your prompt | The original clip |
| Original footage in the result | None, fully regenerated | None | Everything but the face |
Get Better Results
- 1
Let the clip carry the motion
Do not narrate the movement shot by shot. A prompt that describes the action in detail competes with the reference instead of reinforcing it, and motion transfer resolves that conflict by doing neither well.
- 2
Keep faces clear and unobstructed
Identity is read from the character images alone. A well-lit, front-facing photo with the whole face visible transfers far more reliably than a small, angled or partly covered one. If two characters look similar, give them visibly different hair or clothing in their reference images. The model binds each name to whatever distinguishes them, so an unambiguous difference is worth more than a higher resolution photo.
- 3
Shorter references cost less
The reference clip counts towards the billed duration, so a six-second clip with a five-second output costs less than a fifteen-second clip with the same output. Trim the part you actually want before uploading.
Motion Transfer FAQ
Start Creating with Grok Prompts Today
Join thousands of creators using Grok Prompts to generate professional images and videos. Try it free today.