Skip to content
Swap Meet

Free

No. 026Free

talk-to-slides: a talk recording becomes a slide report

Reads a local recording and writes each slide as an image, the text on it, and a short HTML report. Captions let the report cover what was said.

by Lucas Cooper-Bey

About it

talk-to-slides reads a local recording and writes slide images, the text on each slide, data.json and report.html. A captions file adds what was said. With the claude command logged in, the report adds a summary, key moments and takeaways. Use it only on recordings you have the right to process and whose speakers agreed. Limits: Needs Python 3.11 or newer, plus ffmpeg and ffprobe. Slide text uses Apple Vision on macOS, tesseract on Linux, or --ocr none to keep images without text. Without captions it has no speech. transcribe is a separate command and there is no default Whisper address. Without the claude command on PATH, the report has slides and text but no summary. The API backend needs ANTHROPIC_API_KEY. At most 400 frames are read unless you raise the limit. One cleanup call took about 11,000 input and 7,000 output tokens on talks of 6 to 14 minutes. Door 25 of 100 Doors.