Your copy-and-paste prompt
Replace the [brackets] with your details, then paste this into your assistant.
Find the moments worth replaying in my sky footage from [one night or interval]. Use only [recordings I own or an already authorized archive]. The camera time zone and known clock offset are [details or unknown]. Work within [time/resource limit], return at most [five] candidate events, and save new files in [private output folder]. Keep the original recordings unchanged. This is a one-time review, not permission to operate a camera or create a monitoring service. CHECK WHAT WAS ACTUALLY RECORDED Inventory filenames, playable durations, dimensions, frame rates and usable time information. Separate the recording's timestamp from a file's modification date. If clock alignment is uncertain, use filename plus elapsed time and say that the wall-clock time is unknown. Do not infer an exact location from the images. Use sky-only material; if people, neighboring windows or other private ground details are visible, stop and ask for a sky-only crop or export before analyzing them. Make a simple coverage timeline for the requested interval. Show available recordings, missing periods, unreadable files and the portions you actually inspect. Distinguish recorded coverage from analysis coverage. If these are selected clips rather than a continuous night, say so. If the source is a time-lapse, preserve its supplied capture-time mapping or label times as playback offsets; do not treat playback seconds as real event duration. TRY A SMALL SAMPLE, THEN SEARCH WITHIN THE LIMIT Start with one short representative segment and show the proposed method and a few real sample frames. Use available non-generative processing to find changes or candidate motion, then inspect consecutive frames around each proposed event. A supplied detection index may help choose intervals, but its labels and scores are suggestions, not identities or calibrated probabilities. Do not install a model, create credentials or connect a new service for this pass. State the temporal sampling and any downscaling you use. Sparse samples can miss a brief streak entirely, so never turn 'nothing found in sampled frames' into 'nothing crossed the sky.' If available tools cannot inspect enough consecutive frames for the task, return a coverage note and a clearly labeled manual-review plan. Do not invent a sighting to make the result interesting. BUILD A SHORT REVIEW SHEET For each candidate, assign a stable event ID and keep a representative original frame plus a short clip with a few seconds of context before and after, where available. Store annotations separately or offer an unmarked view. Do not generate missing frames, erase objects or enhance a speck into invented detail. Record the source filename, start/end offsets, any trustworthy wall-clock time, the observed appearance and the reason it was shortlisted. Describe what is visible first: for example, a brief streak, a blinking moving point, a fixed light or an uncertain artifact. Keep proposed explanations separate from observations. Do not label a meteor, satellite or aircraft as confirmed from an ambiguous image, and do not identify or track people. An unknown is a useful result. Check neighboring files before counting events: one moving object can cross a file boundary or trigger several detections. Group overlapping fragments only when the timing and motion support it; retain the source references and mark unresolved duplicates. Avoid counting frames as separate sightings. Keep a small rejected/uncertain list with reasons, so I can inspect obvious false positives without browsing the whole night. DELIVER SOMETHING I CAN OPEN Make a compact local HTML contact sheet with thumbnails, event IDs, plain-language notes and links to the exported context clips. Include an editable CSV event index and the recording/analysis coverage note. Use relative local paths that still work after reopening the folder, not expiring signed archive links. Keep a manual review-status field, initially Unreviewed; your shortlist is not my confirmation of an identity. Open the actual sheet and every delivered clip. Check that each thumbnail and clip corresponds to its recorded filename and offsets, that clips play, and that file-boundary fragments are not silently lost or duplicated. Report checks performed, time-mapping uncertainty and inaccessible intervals. A clean sample does not establish detection accuracy or completeness for the night. Return the folder, a one-paragraph viewing guide and the most interesting unresolved question. If nothing earns a shortlist place, return an honest coverage report with no invented highlights. Do not delete or relabel the source archive, train on my footage, publish clips, change retention, install camera software or start recurring jobs without separate approval.
Automatic copying is unavailable. Select the prompt text above and copy it.