An AI developer's account of running Gemma 4 on a MacBook to generate descriptions for a large number of video files is fascinating

SimbaStack shared a workflow for running Gemma 4 locally on a 2021 M1 Max MacBook with 64GB RAM.

The setup turns a year of video footage into Markdown indexes using metadata extraction, frame sampling, GPS lookup, transcription, face recognition, and visual captions.

Tools in the pipeline include LMStudio, ffprobe, ffmpeg, exiftool, Nominatim, WhisperX, and insightface.

The author also open-sourced the configuration on GitHub, showing how personal video archives can be searched without cloud services.