Characters
A character is a name, one to three reference images and a short description. Name it in a generated clip and those images go to the model with the prompt, so the same face comes back in every shot — on Veo, Seedance or MiniMax alike — instead of drifting from scene to scene.

Create a character
With the MCP tool create_character, in one of two ways.
From images you already have — free
Pass one to three image URLs that are media on your account: a generated still, an asset from your bank, or a shared asset from list_commons_assets.
{ "name": "create_character",
"arguments": {
"name": "Ada, the founder",
"refs": ["https://storage.googleapis.com/…/ada-front.png",
"https://storage.googleapis.com/…/ada-three-quarter.png"],
"description": "A woman in her late twenties, platinum blonde bob, grey tee, calm and direct."
} }
Draw the reference — the price of one still
No images yet? Pass generate and PageToVid draws a clean reference portrait (neutral background, even light, the whole subject in frame) for 40 credits, refunded if the drawing fails.
{ "name": "create_character",
"arguments": {
"name": "Sam, the PageToVid host",
"generate": "A friendly man in his early thirties, short dark hair, light stubble, navy crew-neck sweater, at a desk in a bright home office, looking into the camera",
"description": "Sam, a friendly man in his early thirties, talking to camera like a YouTuber."
} }
The drawn portrait lands in your media bank too. If the character is then refused (say, a name longer than 60 characters), the error gives the portrait’s URL: retry with it as refs and you are not charged again.
| Field | Rule |
|---|---|
name | Required, at most 60 characters. Creating a character with an existing name replaces it. |
refs | 1 to 3 image URLs on your account. Three is the model’s own ceiling for reference images. |
generate | Draws the reference instead; ignored when refs is given. |
description | At most 300 characters, added to every clip’s prompt so the words and the picture agree. |
site_url | Optional: attach the character to one of your sites. |
list_characters shows your characters; the REST API has the same: GET / POST /api/v1/characters and DELETE /api/v1/characters/{name} — see REST API.
Use it in a clip
Add data.character to a clip scene, in create_animation or update_storyboard:
{ "visual": "image",
"heading": "Ada at work",
"narration": "Ada used to spend her Mondays on reports.",
"data": { "generate": "Ada at a standing desk, sighing at a spreadsheet, slow push-in, cool office light.",
"clip": true,
"character": "Ada, the founder" } }
It works with any model: add "model": "seedance-2.0-mini" and Seedance receives the same reference images. generate_asset takes character too, to make a single clip or still of the character into your bank. See AI clips for everything else about clip scenes.
Use it as a presenter
A character can also talk in the corner of the frame while the real site is filmed full-screen — the classic screen-tutorial layout:
{ "op": "set_inset", "scene_id": "YOUR_SCENE_ID",
"character": "Sam, the PageToVid host",
"generate": "Sam at his desk, talking to camera. He says: \"Step one: paste a link.\"" }
The full recipe, with a film made that way, is in A presenter in the corner.
Best practices
- A clear, front-facing reference. Face the camera, eyes visible, even light, a plain background. Add a three-quarter view and a full-body shot as the second and third references if you have them.
- Simple, distinctive wardrobe. A navy sweater survives eight shots; a patterned shirt with a necklace does not. Put the wardrobe in the
descriptiontoo. - No text in a reference. A reference with words in it teaches the model to draw words. To build a character from a film you already made, use a scene’s
raw_frame_urlfromget_storyboard— never itsthumbnail_url, which is cut from the finished film with the headline and caption burnt in. - One character per clip. A scene names one
character; two people in one generated shot is where identity drifts first. - Short spoken lines, in quotation marks.
He says: "Two minutes, that's all."— one sentence per clip. A scene with empty narration is held until the character’s last word. - Describe the action, not the face. The references carry the face; the prompt should spend its words on what the character does, where, and how the camera moves.
Limits
- Characters are used in generated clips and stills, which are a paid-plan feature.
- There is no character editor in the web app yet: characters are created over MCP (
create_character) or the REST API (POST /api/v1/characters). Once created, a clip made in the web editor does not name one — useupdate_storyboardto put a character in a scene. - At most three reference images per character.