Sign in to prepare something to watch.
Times refer to the original, unedited file. Word intervals include a small mute buffer. Video skips may expand to nearby keyframes when a copy is prepared.
Nudity labels describe what the detector classified. Buttocks labels do not distinguish male/female, and cleavage is not a separate category.
Expiration is measured from when this copy was prepared. Keeping watched episodes uses their existing cache slots; remove them to allow more episodes.
This report shows original-file cut ranges and the mute ranges applied to the prepared file. It includes file paths, but no account credentials.
This separate test saves lower-confidence candidates from this title only. Candidates never affect filtering or overwrite your normal scan findings.
Old scans did not save rejected candidates. Run this test once to capture them. It uses your current model and sampling rate in 60-second chunks. Candidates are captured down to 10% model confidence. Sparse sampling and model recognition errors can still cause misses.
Enter both values to test only that section. Starting a different section replaces the previous diagnostic results for this title. Normal scan findings stay unchanged.
Slower: tests eight versions of each sampled frame, using 720p decoding and overlapping crops. Requires a section of 120 seconds or less. Results identify which version detected each candidate. Normal library scans are unchanged.
These are individual sampled-frame labels, not confirmed scenes. Scores are model confidence, not measured accuracy. Display changes use saved data and do not rescan. VLC opens the original on the server PC.