Descriptions:
Stephanie Nyarko shows how she built a fully automated video editing workflow inside Claude using a custom skill — no coding required. The system reads a video transcript, determines placement for titles, lower-thirds, callouts, and transitions, generates custom branded graphics via Google’s Imagen model (referred to in the video as “Nano Banana”), and renders the finished edit using FFmpeg, the open-source command-line media tool. An edit that previously consumed three to four hours now runs unattended in the background.
The technical foundation is Claude’s computer use environment (“Cool Work”), which gives Claude the ability to interact with the local filesystem and execute commands. Skills in this context are markdown files (skill.md) that define reusable agent behaviors — Nyarko writes the editing logic once and calls it on demand. Before rendering, Claude presents a structured plan showing the intended edits (intro string, section titles, callout positions) for approval, keeping the human in the loop without requiring manual timeline work.
The live demo edits the first 30 seconds of a prior video about Claude Opus 4.8, with the output showing formatted callouts, automated brand-name corrections, and motion graphics rendered through FFmpeg. Nyarko, a solopreneur who also uses a human editor for longer productions, frames the skill as a complementary tool for shipping more content faster — particularly valuable for solo creators constrained by time. She also mentions having a set of skill packs available for download covering a range of business and personal productivity use cases.
📺 Source: Stephanie Nyarko · Published June 02, 2026
🏷️ Format: Hands On Build
