Turn a video or text into a reusable motion
Last updated: October 15, 2026
Teach Layer a move once, from a reference video or a written description, and play it on any character mesh in Animate mesh.
A trained motion lives in a reference set, like a style or a character. Train it once and the same move plays on every character you animate with it, so a cast shares one cast animation, one victory dance, one hit reaction.
Before you start
- Training needs the Train styles permission in your workspace role.
- To learn the move from footage, put one reference video of it in a reference set: one performer, the whole body in frame, the camera steady. To describe the move in words instead, you need no video. If the set doesn't exist yet, create a reference set first.
- Set the reference set's category to Animation (the category menu above its name in the reference set), so the Animations slot of a 3D form offers it.
- You need a character mesh on a session board to play the move on. It doesn't need a skeleton: the run fits one.
- Training costs Creative Units; the price shows on the Train motion button. See Creative Units.
Steps
-
Open the reference set that holds your reference video. On a session board, double-click its tile.
-
In the 3D section, select Create finetune, then under Base model pick the motion model.
-
Under Method, choose Video to read the move from your clip; the setup confirms that the motion is extracted from your attached reference video. To describe the move instead, choose Text and write a Motion description: the action, its style and its pacing. Leave Model version on its default unless you are matching an older motion.
-
Select Train motion. The motion appears under 3D marked Training. It trains on its own, so you can keep working; you get a notification when it is done.
-
When the row reads Ready to use, go back to the board (select the session's name in the breadcrumb), open Animate mesh and pick the motion model.
-
Select the character on the board and click Source mesh. Then select Animations and add your motion; it lands in the prompt box as a chip.
-
Select Animate. When the result lands on the board, open it: the character plays your move.
Result
The animated character lands on the board as a GLB, rigged and playing the move. In the viewer, the Animation controls pause it and set it to loop, and Show skeleton shows the joints. The motion stays in the reference set under 3D, ready for the next character.
Production tips
- One move per set. Name the set for the move ("Fire mage cast", "Heavy landing"), so the Animations picker reads like a move list.
- Shoot the reference for the extractor. Trim the clip to the move itself, keep the whole body in frame and the camera locked off, and film against a plain background. Loose clothing, props crossing the body and cuts in the clip all make the move harder to read.
- Use the same motion across a cast. Add the one motion set to each character's Animate mesh run, so every character's cast or idle shares its timing.
- Animate once from a video without training. For a move you need only once, Animate mesh takes a clip directly in Source video. Train a motion when the move will be reused.
- Export for the engine. Download the GLB from the viewer, or convert it with Download as.... For presets, rigging options and engine export, see Rig and animate a 3D character.
- Retrain without breaking what you have. Retraining, with a new clip or another model version, adds a new version; the current one keeps working until the new one is ready.
If something goes wrong
- Train motion is greyed out. Hover it for the reason. With Video, the set must hold exactly one video: add one, or remove the extras. With Text, write a motion description.
- Training failed or timed out. Train again; if a run keeps failing, contact support from Get help on the failed run. See Training failed.
- Your motion isn't in the Animations picker. The slot lists Animation sets. Set the reference set's category to Animation, then open the picker again.
- The move plays, but not as filmed. The extractor reads what it sees: a cropped limb, a moving camera or a second person in frame changes the result. Reshoot or trim the clip and retrain.