feat(vcam): use a phone on the network as a virtual camera

SCENES > VCAM pairs a phone through a QR code and turns it into a
handheld camera, after Unreal's Live Link VCAM and Blender's VirtuCamera:

- Android Chrome tracks 6DoF through WebXR immersive-ar; iPhone Safari,
  which has no WebXR AR, tracks rotation through DeviceOrientation, and
  thumb sticks dolly, truck, crane and pan.
- A motion scale (1:1 walk to 1:10000 orbit), smoothing, height and
  horizon locks, hold/clutch and recenter, none of which make the camera
  jump. Focal length and scale-aware depth of field with tap-to-focus.
- The phone shows the app's render as a monitor (newest-frame JPEG pull).
- REC records a take as a version-7 track shot in the selected scene.
- The rig claims the camera through the navigation seam (Cockpit refuses
  it) and holds on any other navigation, scene run or globe input.

Pairing never exposes the app: a loopback-only session request starts a
separate HTTPS listener (VCAM_PORT, default 4174) that serves only the
static phone page and a token-paired relay, and stops after the last
session. Its certificate comes from a local CA built with node:crypto, which
an iPhone can trust once; without it the phone falls back from WebSocket to
HTTPS requests, which iOS allows past a certificate warning.

scripts/qa-vcam.mjs drives it end to end: the real phone page over the
HTTPS fallback, a scripted 6DoF phone, a recorded and replayed take, depth
of field and camera handoff.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015H1iLusTzWpVKitGic2AW8
This commit is contained in:
Claude committed 2026-10-19 04:03:02 +00:00
1 parent c330eb801a
commit 56586a234b
33 files changed
+5472 -6

No files matched your search

+1
View File
@@ -86,6 +86,7 @@ The dev server is a **key broker**: every server-side key above is spendable by
- **Local-only by default.** `./scripts/dev-fresh.sh` (and the Vite config itself) bind to `localhost`, so only your machine can reach the server — and only local names are accepted (`allowedHosts` stays restricted, which also blunts DNS-rebinding tricks). Binding to all interfaces keeps that restriction: IP addresses and `localhost` work, and LAN hostnames must be listed in `GEV_ALLOWED_HOSTS` (for example `GEV_ALLOWED_HOSTS=globe.lan`); suffix and wildcard entries are ignored.
- **LAN exposure is an explicit opt-in**: `HOST=0.0.0.0 ./scripts/dev-fresh.sh`. The launcher prints a prominent warning plus your LAN URL. Understand what opting in means: **every device on that network can drive the proxies and spend your OpenAI / Google / OpenSky / AISStream / TomTom / FIRMS quota** for as long as the server runs. Do this only on networks you trust.
- **The phone camera (SCENES › VCAM) opens a separate, narrow LAN listener — only on request.** Creating a pairing session (a loopback, same-origin request from the app, behind the same gate as `/mcp`) starts an HTTPS listener on `VCAM_PORT` (default `4174`, all interfaces unless `VCAM_HOST` narrows it); it stops two minutes after the last session ends. It serves only the static phone page, a local CA certificate download and the phone side of a relay: never the app, its provider proxies or any key, so pairing a phone does not amount to the `HOST=0.0.0.0` opt-in. A phone joins with the six-character code shown on the desktop (about 30 bits; eight wrong codes rotate it) and receives a per-phone key; the relay forwards bounded JSON (4 KiB, 120/s) and one newest JPEG preview frame, and WebSockets on both ends must come from their own page's origin. The listener answers only IP addresses, `localhost` and `GEV_ALLOWED_HOSTS` names. Its TLS certificate is issued by a CA created in `.gev-cache/vcam/` (private keys owner-only), constrained to server certificates (path length 0); a phone that installs and trusts it trusts certificates from this machine's CA, so remove the profile when you no longer need it. **Limit:** anyone on the network who reads the code off your screen can drive your camera and see its preview until you press END.
- **App-level throttles (on by default):** `GEV_RATELIMIT_OPENAI_PER_MIN` (default `30`) and `GEV_RATELIMIT_GOOGLE_PER_MIN` (default `120`) — the values the Pinokio build already ships — cap the cost-bearing endpoints per client IP per minute (over-limit requests receive a sanitized `429`). Both defaults sit well above what the app itself generates; set either to exactly `0` to run unthrottled. They are **per-IP, process-local, in-memory guards** — they reset on restart, a caller with several addresses gets a bucket per address, and they are **not billing caps**.
- **Docker trusts one address, and only because the port is on loopback.**
Docker delivers the host's own browser to the container from the container