Run a photo or video through the lab.
A battery of forensic pipelines on a single upload: AI object detection, a custom aircraft classifier trained on our own datasets, motion-event tracking, provenance checks against metadata, frame duplication, scene cuts, and audio splices — plus scene context, mundane-object findings, lens-flare optics checks, cloud identification, and astronomy resolution that reconstructs what the sky actually held at the capture time and place. Nothing is stored — the file is deleted as soon as we read it.
Computer-vision models find what is in frame — aircraft, people, vehicles — with confidence and bounding boxes.
A custom model trained on our own aircraft datasets names the closest match, or returns Unknown when nothing fits — the signal that matters for anomalies.
Background subtraction and centroid tracking turn a clip into a timeline of what moved, where, and when.
Metadata, duplicate-frame and scene-cut detection, and audio-splice checks combine into a 0–100 provenance score.
From the capture location and time, astronomy resolution computes which planets, bright stars, the Moon, and satellites were overhead — with azimuth and altitude you can cite.
Scene recognition tells indoor from outdoor and names the setting — an indoor "sky" is a red flag — and flags when a TV or phone screen is in frame.
A dedicated pass hunts for ordinary explanations — birds, kites, balloons, drones — and for outdoor shots a cloud classifier names the cloud genus. Lenticular clouds explain a lot of saucers.
One frame goes to a frontier multimodal AI model for a reasoned, candidate-by-candidate reading — included with every run. Everything else runs entirely on our own servers.
Analysis is a Research feature — it’s how the archive, the map, and every reel stay free for everyone else. This is a heuristic provenance check, not deepfake or AI-generation detection.