Scene documents gain a second kind of explicit move: `move.track`, timed samples of position, orientation and an optional 35 mm-equivalent focal length. Playback, Load, Replay and seek share one sampler, which stretches the recording to the shot's duration (so editing durationSec retimes a take), takes the short arc for longitude/heading/roll and returns recorded samples exactly. Tracks require the ellipsoid altitude reference and are bounded to 10,000 samples. Shots may also carry a `lens`: focal length and a thin-lens depth of field whose distances are divided by a miniature scale, so a 1:1000 rig renders the city like a tabletop model. A new camera-lens owner applies it to the frustum and a depth-of-field post-process stage; shots without a lens restore the default view. SceneDirector accepts the lens as an injected dependency, captures a changed lens with a shot and gains addRecordedShot() for takes. Versions 1-6 import unchanged. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_015H1iLusTzWpVKitGic2AW8
8.0 KiB
Director camera directions
Version 4 scene documents can name geographic anchors and describe an explicit camera move followed by a hold. Existing camera-pose shots still use the ordinary camera flight. Import, playback, Load, Replay and seek accept the new directions; EDIT DETAILS captures anchors and move starts; a graphical path editor remains a later extension.
{
"version": 4,
"scenes": [{
"id": "austin",
"title": "Across Austin",
"anchors": [
{ "id": "west", "title": "West viewpoint", "lat": 30.2672,
"lon": -97.76, "alt": 6000, "altitudeReference": "ellipsoid" },
{ "id": "east", "title": "East viewpoint", "lat": 30.2672,
"lon": -97.73, "alt": 7000, "altitudeReference": "ellipsoid" }
],
"shots": [{
"id": "crossing",
"title": "Cross the city",
"durationSec": 4,
"holdSec": 2,
"camera": { "anchorId": "east", "heading": 10, "pitch": -45, "roll": 0 },
"move": {
"from": { "anchorId": "west", "heading": 350, "pitch": -45, "roll": 0 },
"easing": "cubic-in-out"
},
"visual": { "style": "normal" },
"layers": {}
}]
}]
}
One owner for each authored value
An anchor is a scene-local geographic position, with a stable id, optional
title, latitude, longitude, height and explicit altitude reference. Each scene
accepts at most 1,024 anchors. IDs must be unique within that scene. References to
missing anchors fail validation before replacing a project.
The shot's camera is its destination. It can be an inline pose or an
anchorId plus orientation. The optional move.from uses the same pose/reference
shape. References cannot also carry inline coordinates. They do not form chains.
Heading, pitch and roll belong to each camera pose, so two shots can look in
different directions from the same geographic anchor. Omitted orientation uses
heading 0, pitch −35 and roll 0 degrees.
An explicit move requires move.from, move.easing, durationSec and holdSec.
Its duration is 0.2–86,400 seconds; the hold is 0–86,400 seconds. The destination
is not copied into another field. linear and cubic-in-out are the supported
easings. A shot without move retains the existing ordinary flight, even in a
version-4 file. Its destination may also reference an anchor.
Capture continues to create ordinary camera-pose shots. Update Shot stores the current viewport as an inline destination, preserving an explicit move's start and easing; it does not change a shared anchor and thereby move other shots. Editing imported directions and anchors currently means editing the scene JSON.
Coordinates and altitude
Latitude/longitude use WGS84 degrees. Orientation uses the application's Cesium
heading/pitch/roll convention in degrees: heading rotates from local north,
negative pitch looks down, and zero pitch is horizontal. Camera positions use
meters above the WGS84 ellipsoid, matching Cartesian3.fromDegrees and captured
camera poses. These heights are not mean-sea-level elevations or heights above
the visible surface.
Every anchor and every inline endpoint of an explicit move must say
"altitudeReference": "ellipsoid". Ordinary legacy poses without that field
already use the same reference. Terrain-relative and mean-sea-level references
are rejected; no terrain request, default ground estimate or implicit conversion
is made. Authors must provide ellipsoidal heights with enough clearance for their
scene. The format does not promise collision avoidance or terrain following.
During an explicit move, latitude and altitude interpolate between endpoints. Longitude takes the shortest signed arc across the dateline; heading and roll also take their shortest signed arcs. An exact 180-degree tie takes the negative arc. This is an authored coordinate interpolation, not a geodesic or a camera flight with an automatically raised arc. Endpoints are exact, including zero pitch. Curved paths and look-at targets are later extensions.
Time, seeking and cancellation
Load and Replay of an explicit move start at its authored from pose, regardless
of the current viewport. Play uses that same start. Both live playback and seek
use sampleCameraMove; a seek into the hold uses the exact destination. Backward
seeks do not depend on a prior camera position. Existing scene-pack media/reveal
rules can extend holds as before.
Each active explicit move owns one animation-frame callback. Stop, replacement, import, teardown, a navigation handoff or manual pointer/wheel input revokes that callback and settles the move. Navigation also cancels a run during an authored hold, so the next shot cannot retake the camera. Camera claims still respect Cockpit's refusal. Already queued callbacks cannot move a replacement scene. The timing diagnostic includes the active authored camera animation.
The renderer-independent gods-eye-view/director export provides
resolveCameraPose(scene, camera), resolveCameraMove(scene, shot) and
sampleCameraMove(move, progress). These expect validated/normalized data.
Rendering and frame ownership live in src/scenes/cameraMotion.js; the existing
scene controller supplies camera application and lifecycle cancellation.
Version-1/2/3 projects migrate to version 4 without converting their ordinary shots into explicit moves. IDs, positions, durations, visual settings, packs and source attribution survive. The bloom scale remains independent of the document version; version-3 bloom is never interpreted as the older inverted scale.
Recorded tracks
Version 7 adds a second kind of explicit move: a recorded track, which is what
VCAM takes write. Instead of from and easing, move.track holds
timed samples:
{
"id": "take-1",
"title": "Take 1",
"durationSec": 6.4,
"holdSec": 0,
"camera": { "lat": 30.27, "lon": -97.74, "alt": 950, "heading": 20,
"pitch": -25, "roll": 0, "altitudeReference": "ellipsoid" },
"move": {
"track": {
"altitudeReference": "ellipsoid",
"samples": [
[0, -97.7431, 30.2672, 900, 10, -30, 0, 24],
[6.4, -97.7400, 30.2700, 950, 20, -25, 0, 50]
]
}
},
"lens": { "focalMm": 50, "dof": { "enabled": true, "focusM": 300, "fStop": 2, "scale": 100 } }
}
Each sample is [t, lon, lat, alt, heading, pitch, roll, focalMm?]: seconds from
the start of the take, then the same coordinate, height and orientation
conventions as a pose, then an optional 35 mm-equivalent focal length (4 to
2,000 mm). The first sample is at 0 seconds and times strictly increase. A track
has 2 to 10,000 samples and requires the ellipsoid altitude reference. The shot
camera stays the destination: the final sample, which Load, Capture and the
ordinary flight from the previous shot use.
Playback stretches the recorded time to durationSec, so changing the duration
retimes the take without touching its samples. Between samples, latitude, height,
pitch and focal length interpolate linearly; longitude, heading and roll take
the short arc. A time that lands on a sample returns that sample exactly,
including both ends. Playback, Load, Replay and seek all use
sampleCameraMove, which dispatches to sampleCameraTrack for tracks.
Lens
Version 7 shots may carry an optional lens:
focalMm: 35 mm-equivalent focal length across the frame's longer side (4 to 2,000 mm), applied to the camera's field of view. Cesium's default 60-degree view is about 31 mm.dof: depth of field, withenabled,focusM(focus distance in metres, 0.1 to 10⁸),fStop(0.5 to 64) andscale(0.001 to 10⁷). The blur is a thin-lens circle of confusion computed after dividing distances byscale, so a scale of 1,000 renders the scene as a 1:1000 miniature.
A shot's lens is applied after its visual state. A shot without one restores
the application's field of view and turns depth of field off, so older shots
look as they always did. Capture and Update Shot store the live lens only when
it differs from the default. A track's per-sample focal length overrides
focalMm while the track plays.