AI models

A generated clip — a scene’s own clip or a presenter in the corner — is made by the model the scene names. The choice is per scene, not per account: one beat can use a fast draft model and the next a premium one.

  1. Choosing a model
  2. Available models
  3. What PageToVid asks the model for
  4. Cost and failures

Choosing a model

Set model on an image scene’s data (with clip: true), or on set_inset:

update_storyboard  video_id: …
  set_data  scene 1  data: { "clip": true, "model": "seedance-2.0-mini",
                             "generate": "Cinematic close shot: morning light across a wooden desk…" }

Omit model and the clip is the house clip: Veo 3.1 Fast, 8 seconds at 720p, with sound. Veo 3.1 Standard is used only when a scene names veo-3.1.

list_models returns every model this account can use right now, with its lengths, resolutions and price in credits; include_unavailable: true adds the others with the reason.

Available models

Prices are not repeated here: they are in the generated model price table, which is regenerated from the price registry on every price change.

Model id Family Sound Length Resolutions Status
(omitted) Veo 3.1 Fast (house clip) yes 8 s 720p default
veo-3.1-fast Veo 3.1 Fast yes 4, 6 or 8 s 720p, 1080p, 4K live
veo-3.1 Veo 3.1 Standard yes 4, 6 or 8 s 720p, 1080p, 4K live
seedance-2.0-mini Seedance yes up to 15 s 480p, 720p live, verified
seedance-2.0-fast Seedance yes up to 15 s 480p, 720p available
seedance-2.0 Seedance yes up to 15 s 480p, 720p, 1080p, 4K available
seedance-2.5 Seedance yes up to 30 s 480p, 720p available
minimax-h3 MiniMax yes 4–15 s 768p, 2K live, verified
minimax-h3-max MiniMax yes 5–15 s 480p, 768p live, verified
wan-3.0-prime Wan yes 2–30 s 480p, 720p, 1080p priced, not yet switched on

In a film, Veo clips are always made at 8 seconds; the other families get a length planned from the scene, as below.

What PageToVid asks the model for

You do not set a clip’s length or resolution — the request is planned from the shot:

  • Length: as long as the scene will be on screen (its narration, or its silent hold), within what the model can make. A clip that outlasts its scene costs a moment nobody sees; one shorter than its scene would freeze on its last frame, which is what this avoids.
  • Resolution: the cheapest tier for a presenter in a corner (a quarter of the width never shows the difference), and at least 720p (768p on MiniMax) for a clip that fills the frame.
  • A length the model only advertises is never asked for below the length proven on a live job; if a model refuses a length, it is asked again once at the proven one.
  • A character’s face goes to the model as its reference images, so a named character is the same person on Seedance or MiniMax as on Veo.

Three Seedance 2.0 Mini clips from the film on the home page

Cost and failures

The house clip costs 311 credits on top of the 40-credit render; a named model costs its price in the table, per second at the resolution made. quote_cost prices any request first, for free, and refuses a length or resolution the model does not offer rather than changing it. A clip is refunded if it fails. The render never stops for a failed clip: the scene falls back to its page still and its text, and get_video says so in warnings and ai_report — with the model’s own reason, such as a refusal on content grounds (change the words) or a timeout (try again).