Capture sources, selection rules and video recording
Capture collects stills from a camera, supported screen sharing, a playable local video file or, on desktop, a network camera into a staging gallery. Photo mode offers Manual, Timelapse, Motion and Smart capture. Camera, screen and network-camera sources can record silent clips when supported; watched folders are not implemented.
Choose the source before the capture rule
Choose Camera, Screen, Video file or, in the desktop app, IP camera. Manage sources lets you name reusable inputs and test reachability, connection latency and observed frame dimensions when available. Screen capture opens the system picker for each session and is disabled on mobile and in the native macOS app. Video-file sources support speed and looping; the app keeps a private local copy for later sessions, falling back to file selection if it cannot save or restore that copy. Captured stills retain their recording timestamp. An IP camera source takes a direct RTSP/RTSPS or HTTP(S) stream address or a JPEG snapshot endpoint and decodes it with the bundled FFmpeg on Windows and macOS; its password is kept for the session only. Watched-folder adapters are not implemented.
Nothing enters the dataset until Accept
Open Capture from Upload media, select the source and grant its required permission or choose a playable file. Photos and completed video recordings first appear in the staged session, where they can be inspected or removed. Accept imports the remaining items; Discard all removes the staged session. A recording is finalized before it appears, so leaving, changing camera or changing capture type is blocked while the recorder is active.
Four capture modes
| Mode | Capture rule |
|---|---|
| Manual | Take one photo when the shutter is pressed |
| Timelapse | Take one photo every configured number of seconds |
| Motion | Capture when the frame changes beyond the motion threshold |
| Smart | Capture when the selected model signal says the frame is useful |
Video recording
For a camera or screen source with recording support, select Video under Photo in the left rail. The auto-capture modes disappear because they produce still frames, and the shutter becomes Record. Press it once to start and again to stop; the live REC timer shows the active duration, and the finalized clip joins Collected clips. Resolution and frame rate follow the selected camera capabilities, while the recorder negotiates a supported WebM or MP4 type at runtime. The mobile limits are deliberately lower than desktop limits so a WebView cannot be killed by an oversized in-memory recording.
Camera video is silent. AnnotateIt does not request a microphone stream for Capture, so choosing Video does not introduce a second permission or record nearby conversations.
Smart sources
| Source | What “useful” means |
|---|---|
| Text prompt | The best match is below the uncertainty threshold; a frame with no detected candidate is not treated as uncertainty |
| Active model | The project’s current prediction source is unsure about its best candidate |
| Diversity | The frame is visually unlike the frames already collected in this session |
The Smart signal selector is capability-aware. Mobile omits Text prompt and Diversity; a project without a ready prediction source cannot use that source. Unavailable choices are not silently substituted.
Rate and volume guards
Automatic modes enforce a minimum interval so one event cannot flood the session, and can stop after a configured number of frames. Motion has its own sensitivity threshold. Smart model sources have a confidence threshold; Diversity has a similarity threshold. The live status explains whether capture is armed, paused by a source that is not ready or stopped by the limit.
Optional pre-annotation
Text-prompt and active-model sources can stamp their confident predictions onto a captured still before it reaches the staging gallery. These annotations are produced on the device together with the capture; the camera feed and the image are never uploaded. Review the result after import just as you would any model-assisted annotation.