Show HN: AI search for every photo and every frame of video on macOS
Points and comments are a snapshot, not live.
Show HN: A macOS app for AI-powered search of every photo and video frame using local models.
SCM (Screen Memories) is a local-first macOS app that indexes photos and videos for semantic search. It uses ONNX runtime for vision models (CLIP, SigLIP) and Whisper for dialogue transcription, with OCR via Tesseract. Five search modes cover files, scenes, dialogue, OCR text, and optional LLM chat (Qwen3 or Llama). Videos are segmented by scene detection for per-shot search. The app runs fully offline after an initial model download, with no cloud uploads. It supports watched folders, deduplication, and up to 20 saved search tabs.
What commenters are saying
Commenters noted concerns about processing time: one user with 12,000 videos worried about duration, and a reply estimated keyframe-sampling could take an overnight run. Another recommended scene detection over fixed sampling. One commenter suggested using Apple's Vision framework instead of Tesseract for OCR, claiming it's faster and more accurate. Two camps emerged: those praising the local-first privacy approach, and those questioning the Electron/JavaScript stack versus a native or Rust implementation. Some users expressed interest in facial/pet search features. The thread also debated whether LLM-generated code tends to produce inferior software.