---
title: A software walkthrough that writes its own chapters
summary: Index a screen recording, find every feature you named in the transcript, place a callout at each one, and export YouTube chapters from the result.
lede: Callouts where you pointed, and YouTube chapters from the recording
outcome: design
tools: index_video, search_transcript, sample_frames, list_fonts, author_overlay, render_html, make_captions, render_ffmpeg, export_edit, job_wait, create_project, commit_to_library
works_with: screen-recorder
order: 0.5
---

> Index ~/Movies/walkthrough-take3.mov into GroundSlate. It is a screen recording, nobody is on camera. Then find every point where I name a feature and give me the timecode and what I said. Sample a frame at each one, propose where on screen the callout should sit, and wait for me to confirm the positions. Then build an overlay from that: a callout at each confirmed position and section titles in the lower third in Montserrat, then karaoke captions along the bottom that highlight word by word, and nothing else on screen. Lay it all over the recording at 1920x1080. Then give me YouTube chapters from the points where the interface changed, and file it under a business project called "Docs 2026".

- You narrated the structure while recording, so the transcript is a better outline than one written afterwards.
- Nothing detects what you were pointing at. Your agent looks at a frame from each timecode and proposes an x and y for the callout, and you confirm or correct it before anything is built.
- On a screen recording, shot boundaries fall where the interface changed, and those become the chapters.
- The callouts and titles go on first. The karaoke captions are burned in after, in a separate pass.