Steve Bransford released cairn.js in late August, an MIT licensed library that adds captions and narration to walkable gaussian splat scenes.
WebVTT captions bind to a media element and ride a single timeline. In a walkable scene the visitor's movement decides what plays and when. cairn.js splits the job into three layers. Zones in the scene trigger narration fragments, standard unforked WebVTT times the cues inside each fragment, and a state layer tracks what has already played, holds playback to one voice at a time, and keeps a running transcript. The project treats text as the primary track and audio as the enhancement, so a muted scene still carries the full narration.
The cairn.js core is engine agnostic and ships with HTML Audio and PlayCanvas adapters, WCAG minded overlay styles, and a written spec for other engines to implement. A Python authoring script turns a narration script and Whisper word timings into VTT, cue JSON, and SRT files for Premiere and Resolve.
cairn.js 0.2.0, published to npm on August 27, added persistent queues for zones a visitor crosses in sequence, an interruptWith() call that fades narration for a one off encounter and then resumes it, and resume from the last cue boundary after a page reload. Version 0.2.1 followed a day later as a metadata patch. Bransford describes the 0.2 line as a release candidate and calls the direct PlayCanvas sound slot adapter experimental. The GitHub repository carries no tags or releases, so npm is the only publication channel.
The W3C Immersive Captions Community Group published recommendations for accessible captions in 360 degree video in April 2024, a format that still plays against one master timeline.
Atlanta Space Machine, the reference deployment, links twelve narrated, walkable splat scenes of Atlanta through an aerial view built on Google 3D Tiles, with captions on by default. Each scene carries CC and Start audio controls, and the Mason Mill Ruins scene switches between October and December 2025 captures of the same graffiti covered ruin. Bransford says he captured most scenes with an XGRIDS PortalCam, processed two in LichtFeld Studio, and built all of them on the PlayCanvas Engine.
WebXR presentation and multi language manifests remain on Bransford's list of future work. cairn.js is available on npm under the MIT license.
