Paste a YouTube link into ChatGPT and it reads the transcript, not the picture. crv runs locally: scene-change detection picks the frames that matter, dedup drops the repeats, Whisper transcribes the audio - then any LLM can read the folder. MIT, no ML downloads.
