[Bug]: 3D Orbit camera does not follow the cursor in projects created after the first (Windows 11, 2.1.2 Store) #1397

Search existing issues

• I searched the existing issues. The closest reports are [Bug]: Auto zoom does not follow the cursor #72 (1.x, auto zoom did not follow the cursor, fixed in fix(zoom): auto-placed zoom regions follow the cursor for their whole span #118) and feat(zoom): give the 3D Orbit camera the zoom's focus mode #834 (3D Orbit reads the zoom's focus mode). Neither describes the camera and focus mode being inert in later projects while the same install renders the first project correctly.

Describe the bug

Reported by a user on Windows 11 with the Microsoft Store build of OpenScreen 2.1.2 (translated from Chinese):

The first project (draft) I created after installing worked: the 3D camera followed the mouse cursor. In every project created after that, automatic zooms are still generated, but the 3D camera no longer follows the cursor. I tried uninstalling and reinstalling, and the problem persists.

The user supplied a 23.9 s screen recording of the editor (1272×720, 30 fps) that shows a draft where Auto focus is selected and the 3D camera does not follow the pointer.

What the recording shows

• The draft is a 12.4 s take with no webcam; project format 2560×1284.
• One zoom span on the timeline, starting within the first second and running to the end of the take, at 1.80×. That span is too wide for an automatic zoom (the pass keeps the first 2.5 s and the last 1 s of the take wide, and starts a zoom 0.5 s before its first click), so this is most likely a hand-made zoom the reporter widened to test the camera. It has the timeline's selection highlight, so the inspector is editing this very region.
• Around 0:03.5–0:10 the Zoom inspector is visible: 3D camera = 3D Orbit, Cursor = Visible, and the reporter switches Focus mode Manual → Auto → Manual → Auto. Every click is reflected in the inspector, but the rendered frame never changes.
• The recording does carry editable-cursor telemetry: the rendered cursor is visible in the preview and moves with the page (it travels from the "下载 Windows 版" button near 0:17 to the right-hand "用你的 AI 代理制作演示" button near 0:19).
• The preview stays a flat, fixed 1.8× zoom: no orbital movement while the cursor crosses the frame, no perspective/keystone on the screen, and no re-centring when Focus mode is set to Auto. It renders exactly as if the zoom had no 3D camera and a manual centre focus.

In short, the document/inspector reports rotationPreset = orbit and focusMode = auto, but the rendered frame behaves as if neither reaches the compositor in this project.

The video shows only this draft, so the "automatic zooms are still generated in later projects" half of the report cannot be checked from it; it is repeated here as reported.

Expected behavior

With 3D camera = 3D Orbit and Focus mode = Auto, the camera should follow the recorded cursor (the screen turns toward the side the pointer is on) in the preview and in the export. With Manual, the orbit should be posed by the zoom's focus point and hold there.

To Reproduce

As reported (not yet reproduced by a maintainer on a clean Store install):

1. On Windows 11, install OpenScreen 2.1.2 from the Microsoft Store.
2. Record a take with the Editable cursor. For the reporter, the first project created this way worked.
3. Create more projects/recordings; automatic zooms are still generated in them.
4. Select a zoom and set 3D camera = 3D Orbit, Focus mode = Auto.
5. Play the preview: the camera does not follow the cursor; the zoom renders flat and centred.

OS

Windows

OS Version

Windows 11 (the reporter did not give a build number)

Device Type

Desktop / laptop not specified (Windows 11, x64)

Application version

OpenScreen 2.1.2, Microsoft Store build. About box reports: Electron 41.2.1 · Chromium 146.0.7680.188 · Node 24.14.1 · win32 x64 · store · https://getopenscreen.com/ · © 2025-2026 OpenScreen contributors — MIT License.

Additional context / triage notes

• 3D Orbit is a per-zoom camera, not a project-wide default (technical-documentation/docs: "3D Orbit: a camera that moves with the zoom and follows its focus mode"), and automatic zooms are created flat with Auto focus. So the reporter's expectation that it would carry over into every new project is not what the editor promises — but the recording shows the camera is selected on the affected zoom (3D Orbit in the inspector) and the render still ignores it, so this is not just a misunderstanding of the per-zoom default.
• The same install renders the first project correctly and the reporter says reinstalling does not help. Projects live in user data and are not removed by an uninstall, so this points at per-project state (the saved .openscreen document and/or the per-recording sidecars), not a corrupt installation.
• Both rotationPreset and focusMode are per-zoom fields written by the same inspector and sent in the same scene (buildSceneDescription → native compositor), so their both being inert in the same project looks like one scene/document-side cause rather than two independent 3D bugs.

Useful checks on the reporter's machine:

1. Does the setting survive a reopen? Close the editor and reopen the affected project: does the Zoom inspector still show 3D Orbit / Auto? (If not, saving the camera/focus change is failing for that project.)
2. Does an export show the camera? Export the affected project to MP4 and check whether the 3D orbit appears there. If the export is correct and only the preview is flat, the break is in the preview scene path; if both are flat, the stored document is the suspect.
3. Fresh zoom on the same take: press Z to add a new manual zoom on the same clip, set 3D Orbit + Auto, and play. If the fresh zoom follows but the original one does not, the stored zoom region is the suspect; if neither does, the take/preview is.
4. Project assets: the .openscreen project file, the recording and its .cursor.json sidecar, plus a Save Diagnostics bundle from the tray (which includes the native addon/helper versions). Useful to rule out a recording whose cursor sidecar is not next to the video.

Video: the reporter's 23.9 s screen recording (Cap 2026-10-26 at 18.11.39.mp4) is available; the timestamps above refer to it. It is not attached to this issue yet.