Definitions

A small set of vocabulary recurs throughout the rest of this documentation. Get comfortable with these terms before you start authoring — every other manual assumes them.

Project — the top-level unit of work, persisted as a .prj file. A project hosts one or more scenes, an audio mixer, project-level settings, and (optionally) one or more scripts. Open and save through Composer Desktop; play back through Composer Runtime.
Input — a media source: a video clip, still image, audio clip, capture card, NDI receiver, RTMP listener, web page, or generator. The full catalogue lives in the Inputs manual.
Scene — a stack of layers, composited together using blend modes, transparency, scaling, position and cropping. A project hosts as many scenes as you need; each scene runs in parallel on the same render tick. A scene can use the output of another scene as one of its layers (nested scenes).
Layer — one input placed inside a scene. A layer carries its own transform, audio routing, blend mode, and visibility state.
Operator — a plug-in that modifies video or audio passing through it (a keyer, blur, colour correction, EQ, compressor) or performs an action triggered by media (object detection, speech-to-text, replay capture). Operators attach to layers, to scenes, or to the audio mix. The full catalogue lives in the Operators manual.
Target — an output endpoint: an RTMP or SRT publisher, an NDI sender, a Decklink SDI driver, an ASIO audio device, a file recorder, an HTTP webhook, etc. Targets attach to scenes; a scene can have many targets, all running in parallel from the same composited output. The full catalogue lives in the Targets manual.
Connector — a small adapter that translates an external API call (or a Composer event) into actions on the project. Connectors are the bridge between the Script Engine / external automation and concrete project state.

The diagrams below show how these pieces fit together in increasingly real-world configurations.

Three inputs feeding two scenes; Input 2 is shared between Scene 1 and Scene 2

In the simplest case, a project has a few inputs and a few scenes. Each scene picks the inputs it wants as layers, and the same input can feed multiple scenes simultaneously without duplication.

Four inputs and three scenes — Scene 2 uses Scene 1's output, Scene 3 uses Scene 2's, and Scene 3 attaches one Target

Scenes can be chained: a scene's composited output is itself usable as a layer in another scene. This is the foundation of the multi-brand pattern — a base scene captures the live action, and downstream branded scenes layer their own graphics, lower thirds and overlays on top, each shipping to its own target.

A scene with two inputs, each running three operators; the scene's output is sent to two targets

Operators stack on layers (or on the scene as a whole), and a single scene can drive multiple targets in parallel — useful when the same composited program needs to go to a recorder, a streaming endpoint, and an SDI feed at the same time.