A creator’s tool · Find existing material

Forgot the filename? Search for a picture or a moment in a video

SCM indexes Mac photos and videos for content search, timecoded clips, visible text and dialogue. Follow the maker’s demos and setup conditions.

Based on the maker’s October 4 revision, video code and public demos. BitShovel has not installed the app or tested Chinese retrieval or a private library.

SCM Files mode returns music-related images, with the query and selected model visible above the grid.
Complete frame at about 7.5 seconds in the maker’s Files demo, using SigLIP-2-B/16. The README default is CLIP; this example uses a different model. The query and library belong to the maker’s demo. · Open full-size image
In this article4 chapters

中文

Remembered content can be the starting point

SCM, or Screen Memories, adds content search to existing Mac photos and videos. In the maker’s Files demo, music returns players, headphones and album images. This mode combines visual similarity with filename and phrase signals. The result is a file you can open, even when its original name has been forgotten.

Sources and further reading
  • SCM features and requirementsMaker’s documentation · pinned 8dc1037Search modes, model downloads, macOS installation, copied media and default transcription models.

中文

Find a video and jump to a moment

Scenes results carry timecodes. The code detects shot boundaries, selects segments within a limit and indexes each segment’s midpoint frame. Without detected cuts, it uses equal intervals. Opening a hit seeks to its position. A remembered image can therefore lead to a moment, while retrieval still uses representative frames rather than an explanation of every frame.

The maker’s SCM demo opens a film clip and displays Jumped to scene 2:38 above the player.
Original frame at about 4.5 seconds in the maker’s Scenes demo. The displayed 2:38 is a time in the opened film, not the GIF’s playback time. · Open full-size image
Sources and further reading

中文

Text, dialogue and chat use different evidence

OCR searches recognized visible text; Dialogue searches Whisper transcripts. The default is tiny.en, with base.en available, both English models. Chinese OCR packs do not establish Chinese transcription quality. Optional Ask mode answers from extracted dialogue, OCR and filename evidence, with citations you can return to and inspect.

Sources and further reading
  • SCM features and requirementsMaker’s documentation · pinned 8dc1037Search modes, model downloads, macOS installation, copied media and default transcription models.

中文

Leave room for models and copied media

The documented Homebrew install targets Apple Silicon and macOS 12 or later. Default CLIP weights initially download about 435 MB; transcription and optional chat need their own models. Imported media is copied into the managed library, alongside indexes, thumbnails and segments. The maker describes inference on the Mac. Start with familiar material and inspect whether the results fit your retrieval needs.

Sources and further reading
  • SCM features and requirementsMaker’s documentation · pinned 8dc1037Search modes, model downloads, macOS installation, copied media and default transcription models.

See how these changes connect