Product
FrameQuery 1.1: Tune Face Detection to Find More People
FrameQuery 1.1 lets you tune how hard face detection looks, check a frame every second instead of one per scene, and fixes a startup failure.
FrameQuery 1.1 is out. Face detection is now something you control: a sensitivity slider, optional per-frame checking that finds people who never appear in a scene's main frame, and individual thresholds if you want them. It also fixes a startup failure that could leave the app unable to open, and makes face and visual analysis finish when they are actually done rather than waiting for transcription.
All face and voice processing still runs entirely on your machine. Nothing biometric is uploaded.
Find the people your wide shots were missing
Face detection used to search exactly one frame per scene, and that frame is 960×540. A face had to fill about 4% of the frame width to register — roughly a chest-up shot. Anyone in a two-shot, on the far side of a court, or simply turned away in that one frame was invisible.
The new Detection tuning panel in the People tab gives you three settings — Strict, Balanced, Open — that move confidence, minimum face size and head-angle tolerance together. Balanced is exactly what 1.0 did, so nothing changes unless you ask it to. Open roughly halves the smallest detectable face and admits profiles.
Turn on Advanced and every threshold is individually adjustable, each with a plain explanation of what it trades away. Sharpness, face quality, face-shape tolerance, head turn and tilt limits are all there. Every value is bounds-checked, so no setting can be pushed somewhere that breaks detection.
Tightening a setting now applies to your existing library immediately, with no re-processing — FrameQuery re-filters what it already knows. Only loosening a setting needs a re-process, and the panel tells you which case you are in instead of always sending you to the slow path.
Check a frame every second, not one per scene
Switch on Check more frames for faces and extraction saves roughly one frame per second alongside its normal work, at full source resolution instead of the downscaled scene frame. For an interview where someone turns away during the one frame a scene happened to keep, this is the difference between finding them and not.
Measured on real footage, it adds about 4% to extraction time. Frames that show no new faces are deleted automatically, so a subject sitting still for a minute does not leave sixty near-identical stills on your disk. Face analysis itself runs many more times per video, so it is off by default — turn it on when recall matters more than speed.
These extra frames never leave your computer. They are not uploaded, and they are not part of what FrameQuery sends for cloud analysis.
Already processed a library? Check existing videos re-reads those source files to produce the same frames. It uses no quota, runs at the same concurrency as normal processing, and picks up where it left off if interrupted. Videos on drives that are not connected are skipped and stay queued for later.
Progress that tells you the truth
Face and visual analysis now start as soon as the cloud finishes analysing scenes, instead of waiting for transcription. On a long interview that is minutes earlier.
Scene-text and transcript embeddings were a single step that could not begin until both arrived; they are now separate, and scene embeddings run as soon as the scenes do. The job list has a step for each, so a spinner stops meaning "something in this group is still going".
The "Processed" step is gone. It sat in the middle of the list and claimed to be finished while the job was plainly still processing.
What's fixed
- The app could fail to launch with no explanation after moving your data folder to an external drive or NAS. If that folder cannot be reached, FrameQuery now says so, names the location, and offers to start from the default one — your existing data is left untouched.
- A crash now writes a report you can send us. Previously a crash left nothing behind at all.
- FFmpeg downloads that failed could retry endlessly and stall processing. Downloads now time out, back off between attempts, and report the actual reason instead of a generic failure.
- An interrupted FFmpeg update could delete the working copy and leave the app with none.
- Face and voice analysis no longer appear to run until transcription finishes.
- Fixed bug reports are now grouped by version in the logo menu, with the full history in one place.
- Job progress steps are translated in German, Spanish and French — they were English-only.
How to update
FrameQuery checks for updates on launch and offers the download when one is available. Nothing re-indexes and nothing is re-processed; your projects, tags, people and search history are unchanged. It is a free update for all users, on both macOS and Windows.