AutoMovie API
    Preparing search index...

    Interface IAutoMovieCameraAction

    The camera is an actor; this is how it moves. A list of camera actions composes a move that changes mid-shot ("follow the charge, then hold static on the fall"), which the engine realizes against the target as the shot's camera-node clip (cameraMotion).

    performShot requires the actor to be exactly one staged camera and allows one live camera per shot. Its moves may not share a start or overlap; an "auto" duration runs until the next move starts, or to the shot end. The on target and any focus must resolve to a point (node, live bone, point, or group), with formation groups allowed for lens targets. compileCameraMove owns the path each move takes.

    interface IAutoMovieCameraAction {
        actor: string | string[];
        duration: number | "auto";
        focalLength?: number;
        focus?: IAutoMovieActionTarget;
        focusPull?: { duration: number; start: number; to: IAutoMovieActionTarget };
        framing: "wide" | "full" | "medium" | "close";
        lookahead?: number;
        move: "static" | "follow" | "orbit" | "push-in" | "truck" | "whip";
        on: IAutoMovieActionTarget;
        physicsIntent?: string;
        region?: AutoMovieBodyRegion;
        repeat?: number;
        start: number;
        verb: "frame";
    }

    Hierarchy (View Summary)

    Index
    actor: string | string[]

    The scene-node id or ids performing this action, reusing ids from staging. A list applies the same verb to several actors in unison (a chorus line, a crowd, synchronised dancers) instead of repeating the action per actor. Every id must be a staged scene node, and a list must be non-empty with no duplicates. Only a frame action names a camera, and it names exactly one as a string.

    duration: number | "auto"

    Length in seconds, finite and > 0 with start + duration inside the shot, or "auto" to let the engine pick a natural duration (a stride cadence, a punch's snap, a projectile's flight time). The reference synthesizer sizes an "auto" clip itself, while performShot's overlap check and an attachTo treat the span as running to the shot end, and a frame as running until the next camera move. An "auto" action may not start exactly at the shot end, which leaves it no span.

    focalLength?: number

    Lens intent in millimetres, full-frame equivalent (24 wide, 50 normal, 85 portrait), finite and > 0. Structural intent only: it is copied into shot.cameraIntent, while the scene camera's fovY stays the geometric truth and the solve never reads it. Omit when the lens is not a directorial choice.

    The lens subject, which may differ from the framed subject on. performShot records the world point in shot.cameraIntent at the shot opening, or at the pull's first mark when focusPull is present.

    focusPull?: { duration: number; start: number; to: IAutoMovieActionTarget }

    Timed focus pull from the explicit focus subject to another lens subject. The frame action's interval contains both marks. A downstream host can read these lens marks independently of the camera path.

    Type Declaration

    • duration: number

      Finite seconds from the first mark to the second, strictly positive.

    • start: number

      Finite seconds from the frame action's start, at least zero.

    • to: IAutoMovieActionTarget

      Positional lens subject at the second mark.

    framing: "wide" | "full" | "medium" | "close"

    How tight the framing is.

    lookahead?: number

    Finite nonnegative seconds of aim lookahead for a "follow" move. At a camera key, translation follows the subject's current position and rotation aims at the subject this far ahead, capped at the shot end. Omission or zero aims at the current position. The frame action gate rejects this field on other moves.

    move: "static" | "follow" | "orbit" | "push-in" | "truck" | "whip"

    How the camera behaves over this span.

    What it frames and tracks.

    physicsIntent?: string

    Free-text acknowledgment of a deliberate physical implausibility, such as "defies-gravity" or "intentional-clip", meant to stop the engine's warning-severity plausibility feedback (see IAutoMovieConstraintViolation.severity) from re-firing on this action. No engine or production path reads this field: the plausibility detectors take their own physicsIntent argument, and no shot pipeline calls them, so setting it changes nothing today.

    The body region this action drives (AutoMovieBodyRegion). Actions whose masked content is disjoint compose concurrently (walk while waving while looking), and actions sharing content take turns. Omitted, the engine's actionRegion infers it from the verb and, for gestures, the kind: locomote is fullBody, because the shipped gaits drive hips, knees, and contralateral arms; gesture and reach are upperBody; lookAt and the nod and shake gestures are head; emote is face; the bow, crouch, kick, stagger, jump, and draw gestures and every other verb are fullBody. Overlap is judged on the content surviving those masks (root, bones, expression), so a fullBody gait still layers with a disjoint head-only lookAt. The engine masks the synthesized clip to the chosen region and performShot reports content the mask drops, so override only when the natural region is wrong for the staging. Camera (frame) and attachTo actions ignore it.

    repeat?: number

    Play the synthesized motion this many times back to back, a positive integer defaulting to 1: a step repeated on the count, an idle sway. compilePerformance concatenates the synthesized clip that many times, so the motion runs repeat clip lengths, while performShot's span and overlap checks read only start + duration.

    start: number

    Seconds into the shot when it begins, within [0, shot duration].

    verb: "frame"

    Selects camera framing as the action family.