AI motion transfer: upload a clip, keep the moves, recast it with your product
AI motion transfer on CastTake works like this: give it a reference video and it keeps the motion and the camera while a new hero and your product take the frame. The first 3-second motion preview is free, then it is one clip at a time from $4.99, no subscription.
- Input: a 4 to 15 second reference clip, up to 50 MB.
- Output: up to 15 seconds at 720p or 1080p.
- Price: one-off per clip. Compared with a monthly plan in our Higgsfield comparison.
1. Drop the reference clip you want to recast
Free 3-second motion preview, one per account. Nothing is uploaded from this page: you sign in, then pick the same file in the tool.
What AI motion transfer does
Motion transfer borrows the movement from one video and applies it to a different subject. The reference could be a dance, a person lifting a can to their mouth, a walk down a street or a camera push-in. You keep the choreography and the timing, and you change who is doing it, where, and what they are holding.
People search for the same thing as character swap AI video, video recast or motion control. For a seller the useful version is product placement: take a reel format that already works, put your own product in the hero's hand, and see whether it still works before paying anyone to shoot it.
How it works, step by step
- 01
Upload a reference clip
MP4, MOV or WebM, up to 50 MB. Pick the window you want recast: at least 4 seconds, and the finished clip tops out at 15 seconds including the 2-second opening hook if you add one.
- 02
Say who and what
Describe the hero in the clip, then give your own hero as a look or as text. Add a product photo if you want the product in the shot; a hook (drinks it, unboxes it, shows it) needs one.
- 03
Check the free preview
One per account. It shows the first 3 seconds of the motion we extracted with your hero placed in it, in about 3 minutes, so you can see the movement carried over before paying.
- 04
Pay per clip and render
Preparing the full clip takes about 2 minutes and the render is usually 5 to 8 minutes. You can leave the page; we email you. A failed render is refunded.
Underneath, the pipeline has three stages:
- A depth pass turns your reference video into a movement map: a grey-box stand-in for the shot that carries the motion and the camera.
- A scene plate is generated for the setting you describe, with your hero and product images as references.
- The Seedance video model renders the final clip from the movement map, the plate and the references.
The movement map is why the result follows the reference closely: the video model is not asked to invent the movement, only to draw it. It is also why a clip where the hero is hard to find is refused up front instead of rendered badly.
What one clip costs
AI motion transfer here is one payment per clip, priced by output length. Nothing renews. These figures are read from the same price table checkout uses.
| Output length | 720p | 1080p |
|---|---|---|
| Up to 8 seconds | $4.99 | $6.99 |
| Up to 12 seconds | $5.99 | $8.99 |
| Up to 15 seconds | $6.99 | $10.99 |
The free 3-second motion preview is one per account. Members with credits can pay from their balance instead; every rate is on the pricing page. If you would rather start from a product photo than from a reference clip, the AI product video generator does that.
Who it is for
AI motion transfer suits some jobs much better than others.
Sellers with a product to place
You have a drink, a bottle, a gadget or a bag and a reel format that already works for someone else. Recasting it with your product in the frame is the fast way to test the idea.
Ad testers who need volume
Because you pay per clip, a month of ten tests costs ten clips, not a plan sized for a busy month.
Not for these
A recast of a real person you must keep recognisable, a fast multi-cut edit, or a hero who has to match a specific real face. See the limits below and the comparison with Higgsfield.
Limits worth knowing first
- No real faces as the hero. The video model refuses reference photos that contain a real face. Describe the hero in text or use a faceless or back-view photo.
- Multi-cut sources are hard. We are built for continuous shots. On an 8 second multi-cut fashion reel, Higgsfield Genjutsu kept every cut and recast the model consistently in our test on 28 September 2026; we did not. The Higgsfield alternative comparison has the details.
- The hero must stay findable. If we cannot find it in at least 60% of the frames, nothing is rendered and you keep any credits you paid.
- Sign-in is required. The free preview and the paid render both need an account.
Samples: what went in, what came out
These AI motion transfer clips were already made on CastTake; none is a new render for this page. Each caption says what the input was and what came out. Tap a clip to play it.
Questions about motion transfer
What is AI motion transfer?
It is a way of making a video by borrowing the movement from one clip and applying it to a different subject. You supply a reference video, we work out how the body and the camera move through it, and the video model draws a new subject doing that same movement in a scene you describe. It is also searched for as character swap AI video or AI video recast. The point for an advertiser is that you get a proven piece of choreography or a proven hook without filming it.
What do I need before I start?
A reference clip of 4 to 15 seconds as an MP4, MOV or WebM under 50 MB, a description of the hero in the clip (for example "samurai in a red robe"), and either a look for your own hero or a text description of one. A product photo is optional, except when you pick one of the two-second opening hooks (drinks it, unboxes it, shows it), which need one.
Can I upload a photo of a real person as my hero?
No, and this is the limit most likely to surprise you. The video model we render with refuses reference photos that contain a real face, so a selfie or a model photo will not work as the hero. Describe the hero in text, or use a photo where no face is visible, such as a back view or a character illustration.
What does the free preview include?
One per account: a motion preview of the first 3 seconds of your clip, which shows the motion we extracted and confirms the hero was found. It takes about 3 minutes. It is not a finished render; the finished video is the paid step.
How much does a finished clip cost?
One-off, no subscription: at 1080p $6.99 up to 8 seconds, $8.99 up to 12 seconds and $10.99 up to 15 seconds. At 720p the same lengths are $4.99, $5.99 and $6.99. A render that fails is refunded.
How long does it take?
The free preview takes about 3 minutes. After you pay, preparing your full clip takes about 2 minutes and the render is usually 5 to 8 minutes. You can close the page; we email you when it is ready.
Does it keep the cuts in my reference video?
It is built for a continuous shot. A source with several cuts is harder for us, and in our own side-by-side test on an 8 second multi-cut fashion reel Higgsfield Genjutsu kept every cut where we did not. If your reference is a fast-cut edit, read the comparison before paying.
What happens if my hero is not found in the clip?
If we cannot find the hero clearly in at least 60% of the frames, nothing is rendered and any credits you paid are kept. Try a clearer description of the hero or a clip where the subject stays in frame.
Is this only for dances?
No. Dances are the easiest thing to demonstrate because the movement is obvious, but any action with one clear subject works: a product being drunk, unboxed or held, a walk, a reaction, a trick shot. The optional two-second hooks exist for the product cases.
Ready to try a clip?
Sign in, pick your reference video and get the free 3-second motion preview before you pay for anything.