Introducing AI Captions: Local, Private, and Lightning Fast

Captions make your screen recordings accessible, searchable, and more engaging. But until now, adding captions meant either typing them out manually or uploading your audio to a third-party service and hoping for the best. We wanted to change that.

Today we are shipping AI Captions in One Rec, powered by OpenAI's Whisper model running entirely on your local machine. There is no cloud upload, no API key required, and no subscription. Your audio never leaves your computer, which means your private meetings, internal demos, and confidential product walkthroughs stay exactly where they belong.

How It Works

When you enable captions in the editor, One Rec downloads a compact Whisper model (around 150 MB for the base variant) and runs inference using your machine's GPU or CPU. On Apple Silicon, we leverage Core ML for hardware-accelerated transcription that processes a typical five-minute recording in under thirty seconds. The result is a time-aligned transcript that maps each word to its exact position in the video timeline.

Language Support and Accuracy

Whisper supports over 17 languages out of the box, and we have tested extensively with English, French, German, Spanish, Japanese, and Mandarin. Accuracy varies depending on audio quality, but clean microphone recordings consistently produce near-perfect results. Background music and system sounds are filtered out automatically before transcription begins.

What Comes Next

We are already working on speaker diarization so that multi-person recordings can label who said what. We are also exploring real-time captioning during live streams. If you have feedback on the captioning experience, drop us a note in the community forum — your input directly shapes our roadmap.