SpatialCapture: Agent Camera

Capture reality as a documented ZIP that external AI and Blender agents can inspect: video, camera poses, motion, GPS and optional LiDAR depth.

Turn an iPhone or iPad into a camera for agents.

SpatialCapture records the visual and sensor evidence that external AI tools, Blender scripts and reconstruction workflows need to understand a real scene. Make one take, then share one documented ZIP containing video, calibrated camera movement, timestamps and the sensors you selected.

WORKS WITH OR WITHOUT LIDAR Every supported ARKit device records video and a calibrated camera pose for each accepted frame. A device with a LiDAR Scanner also records measured scene depth and confidence. On a standard iPhone or iPad, the same workflow produces a pose-and-sensor package without invented depth.

CAPTURE • Select the physical 0.5× ultra-wide or 1× wide camera when both are available • Video with optional microphone audio • Calibrated camera positions, orientations, intrinsics and timestamps • Optional gyroscope, accelerometer, gravity and device orientation • Optional GPS coordinates, altitude, speed and accuracy when available • LiDAR scene depth, confidence and matched color references when supported

EXPORT FOR AGENTS AND BLENDER Export a ZIP containing the movie, frame-by-frame JSONL metadata, sensor streams, a readable manifest, a README and a Blender Python importer. LiDAR takes also include lossless depth buffers and color reference images. Recordings remain available in the Files app.

The bundled Blender importer creates an animated camera from every take. For LiDAR captures, it also creates measured surface patches. Use the package as source evidence for manual modeling or with an external reconstruction agent you choose.

PRIVATE BY DESIGN No account, subscription, advertising or automatic upload. GPS is off by default. Everything stays on your device until you use the system share sheet. SpatialCapture has no built-in AI service and does not send recordings to a developer server.

FOR BEST RESULTS The app starts with the 0.5× ultra-wide camera when ARKit supports it, giving agents more scene context in each frame. You can switch to 1× before recording. Move slowly around still subjects and capture several angles. Pose-only captures do not contain measured geometry. LiDAR can miss glass, reflections, distant surfaces and hidden areas. GPS provides approximate geographic context; it does not register the ARKit world to a map.

Requires an ARKit world-tracking iPhone or iPad with iOS/iPadOS 26 or later. The recorder uses an available HD format, preferably 30 FPS. It does not record 120/240 FPS slow motion or create a finished 3D model on the device.

Privacy: https://www.carspecslab.com/docs/spatialcapture-privacy.html Support: https://www.carspecslab.com/docs/spatialcapture-support.html