Skip to content

Models, modes and providers

A model does the generation. A mode specifies the operation and the inputs it accepts. A provider is the service through which the model runs. These are separate choices in the underlying system; the studio presents the choices relevant to your task.

“Make a clip from this opening image” and “change the background of this existing clip” need different operations. Choosing a familiar model name first does not establish that it accepts the inputs you have.

In Video (Видео), choose the intended task, then review the models offered for it. The input ports and warnings explain what the selected combination can use. A wire remaining on the graph does not prove that its file will reach the model.

Term Meaning What to check
Model family and version The named generation model and its release, such as Seedance 2.0 or Seedance 2.5. Use the guide for the actual version; capabilities do not automatically carry between releases.
Mode An operation such as generation from a first frame, reference generation, editing or extension. Required inputs, accepted media, duration and restrictions for that operation.
Task or role The result you are trying to achieve in the selector. Some roles can be implemented in different ways by different models.
Quality or resolution Settings within the selected operation. Available values, estimated cost and whether changing them changes input support.
Provider The service carrying out the model request. Studio availability is the contract you can use; a capability advertised elsewhere may not be connected here.

Shoot from frames (Снять по кадрам) is a task, not one universal model mode. A model with timed keyframes can place images at specified moments. Other models receive a single storyboard sheet as a reference. Kling’s sheet route also needs a separate starting frame.

Those approaches provide different control. Read frames, sheets and keyframes before preparing the images.

Check the combination you will actually run

Section titled “Check the combination you will actually run”

Through MCP, start with describe_studio and use model_advice for the chosen image or video model. Use price_quote for the actual operation and parameters. A family name, default price or successful render on another mode is insufficient evidence for this run.

Model guides distinguish what is connected from what has been tested. “Supported” means the route accepts the operation; it does not promise that every prompt will produce the intended result. A change of version, mode or quality can require a new check.

The provider can depend on the operation. For example, MiniMax H3 replaces a hero in a clip through a different service address than its other modes, and that address accepts only that operation; it is the same model in your project, not a second one to choose. The cost and limits shown for the chosen role are the ones to rely on.

The studio does not expose its internal provider routing as a user setting for every model. MiniMax H3 is the exception: its card has a Directly at MiniMax checkbox, off for new nodes, which moves the same operation from fal to MiniMax and changes the price and the resolution list; it is locked when only one provider carries the operation. Public model names and model authors are sufficient for selecting documented workflows; do not build a workflow around an assumed vendor endpoint.

Continue with the node reference or Seedance 2 guide.

Last checked: 23 October 2026. This page explains studio choices; it is not a vendor capability or price guarantee.