Keep a character on-model in a video

Last updated: October 11, 2026

Put your own character, prop or location into a generated clip by attaching its reference set, so the clip shows the design you approved rather than one the model invents. Use it for trailers, ads and animation tests that feature a recurring character.

Before you start

  • You need a reference set for the subject: a character, object or scene set with a few clean images of it. See Create a reference set.
  • Open the session you want the clip in. Results land on that session's board.
  • How closely the clip follows the set depends on the model. Models that accept reference images use the set's images directly; others use the set's description. The section "How the set reaches the model" below explains the difference.

Steps

  1. In the side panel to the right of the board, under Tools, click Create video. If the panel shows a tool instead of the menu, click the back arrow at the top left of the panel first.

  2. In the prompt, type @ and the start of the reference set's name, and pick it from the suggestions. It appears as a chip. Then describe what the character does, for example: @Frontier Warden turns a full circle on the spot, even lighting, neutral plate. You can also click Reference sets above the prompt and add the set from there.

  3. To direct how the character moves, click Performance and add a preset, such as Slow Turn. Camera, Staging and Look work the same way; see Direct a video with presets.

  4. Check Aspect ratio and Duration, then click Create video at the bottom of the panel.

  5. Wait while the clip generates. A tile appears on the board straight away and shows its progress.

Result

The finished clip replaces the progress tile on the session's board. Open it to check the character against its reference set, or right-click the tile and choose Download.

How the set reaches the model

Layer turns the reference set into whatever the selected model can take, in this order:

  1. A model that takes reference images (or elements) receives the set's images as references and keeps the subject's design from them. A model takes a limited number of reference images, so attach the sets the shot needs and no more.
  2. A model without reference inputs receives the set's written description in the prompt. A character set also opens the clip on its first image when First frame is empty and the model takes one.

The set's first image is its portrait: the one used as an opening frame, and the only one some models read. Put the clearest full view of the subject first in the set.

Production tips

  • Choose a model with reference inputs for on-model work. Models that take reference images keep a character far closer than those that work from text. Compare a couple on the same prompt before a batch.
  • Two characters, two sets. Mention each set where it acts in the sentence (@Frontier Warden blocks @Ash Raider's strike), so the model knows who does what.
  • Curate the set for motion. A front, side and back view helps a turntable or a walk; a set of near-identical portraits gives the model nothing to turn to.
  • Combine with a first frame. For the tightest result, put an approved render of the character in First frame and keep the reference set attached. The frame fixes the opening pose and the set holds the design as it moves.
  • Add direction with presets. Performance presets such as Hero Stance or Held Gaze direct the acting without crowding the prompt.

If something goes wrong

  • The character drifts off-model. The model may be using only the set's description. Pick a model that takes reference images, add an approved still as First frame, simplify the action and shorten the clip.
  • The set is not in the @ suggestions. Suggestions list only sets the selected model can use. Click Reference sets above the prompt and search for it there; if the session is in a project, turn off Only in your project to see sets from the rest of the workspace.
  • Two characters merge or swap. Mention each set at the point in the sentence where it acts, and keep their designs distinct (silhouette, colour). Fewer sets per clip helps.

Related