Turn an expression sheet into talking portraits

Last updated: October 12, 2026

Give a character's portrait the expressions a conversation needs, then make a portrait speak a line of dialogue, for a narrative artist building dialogue boxes, visual-novel scenes or cutscene close-ups.

What you'll make

A set of head-and-shoulders portraits of one character, one per expression (neutral, happy, angry, surprised and so on), all at the same size and framing so they swap in place in the dialogue box. For voiced lines, a short square video of a portrait speaking the line, with the audio in the file, for a cutscene or a talking-head dialogue box.

Before you start

  • The character's approved neutral portrait: head and shoulders, facing the camera, on a plain background.
  • The list of expressions the script needs, and the voiced lines, word for word.
  • For each voiced line, the audio: a voice-over you recorded, or a line you make with Create speech (see Prompt Gemini for text-to-speech). Keep each line to a few seconds.
  • Every run spends Creative Units, and the cost depends on the model; a lip-synced video costs more than an image. See Creative Units.

Bring the portrait in

  1. Open a session and press Ctrl+K (Cmd+K on a Mac), then choose Add asset….

  2. Select the neutral portrait and confirm.

  3. Hover it on the board and select View to open it full-size.

Make the expressions

Each expression is a prompt edit of the neutral portrait. See Edit an image with a prompt.

  1. In the prompt bar under the image, describe the expression and what stays: "Give her a determined look with a slight, confident smile. Keep her face, hair and outfit."

  2. Select Edit image. The new portrait joins the version strip on the left of the viewer.

For each further expression, select the neutral portrait in the strip first, so every face starts from the same one. To get a whole sheet in one go, ask the agent for an expression sheet with the portrait attached, or pick Change expression under With the agent in the viewer's tools; see Work with the Layer agent.

Make it speak

Lip sync image turns a still portrait into a video of the character speaking an audio clip.

  1. With the portrait you want on the stage, open Lip sync image from the tools beside it.

  2. Select Audio, pick the line's recording in the picker, and confirm with Select 1 asset.

  3. Select Lip sync. The talking clip joins the version strip, and the viewer plays it when it finishes.

Play the result through with sound and watch the mouth on hard consonants (p, b, m). If they slip, try a cleaner recording or a shorter line.

Hand it off

Download the portraits as PNG files and the talking clips as MP4 files. In a dialogue system, swap portraits by expression tag; for voiced lines, play the clip in place of the still while the line runs, and fall back to the still when it ends. For Unity, import clips as a VideoClip and play them on a RawImage through a Video Player; in Unreal, use a File Media Source and Media Player; in Godot, convert to Ogg Theora (.ogv), which its VideoStreamPlayer plays.

Scale it up

  • Make the whole cast's sheets the same way, from one neutral portrait per character, at one framing, so every dialogue box lines up.
  • Train a reference set on the character so new expressions and new scenes stay on-model. See Reference sets.
  • For many voiced lines, make the speech and the lip sync in a workflow, one run per line. See Build and publish a workflow.
  • For a full-body cutscene shot rather than a portrait, see Keep a character on-model in a video.

If something goes wrong

  • The character changed along with the expression. The prompt did not say what to keep, or it was made from another expression. Name the face, hair and outfit, and edit from the neutral portrait.
  • The lips do not match the words. Background noise or music in the audio confuses the timing. Use a clean voice-only recording.
  • The face warps or the head drifts out of frame. The portrait is not facing the camera, or the face is small in the frame. Start from a front-facing head-and-shoulders portrait.
  • A run failed or never finished. See Fix a failed or stuck generation.

Related