Key Takeaways

  • Brooke Lacey replaced conventional video editors like CapCut and Adobe Premiere with a terminal-driven script running Claude Code, FFmpeg, and Whisper.
  • The pipeline processes raw video files locally on a Mac in roughly 15 seconds without uploading gigabytes of footage to third-party cloud servers.
  • The workflow treats video editing as a structured text problem, using a tab-separated values (TSV) file where each row dictates timestamps, crops, and assets.
  • Clean comedic pacing requires pairing basic silence detection with Whisper word-level timing, because neither tool alone cuts tightly enough for short-form retention.
  • Lacey formalizes this entire automated production loop as Lacey's 7-Step Headless AI Video Editing Pipeline.

The Lacey's 7-Step Headless AI Video Editing Pipeline

Step 1: Speech and Edge Detection

Pair silence detection with Whisper word timing to mark speech boundaries. Neither tool alone cuts cleanly; Whisper sharpens the edge to hit immediately after the sound of the word ends.

Step 2: Face Tracking and Shot Sizing

Track facial position on every single frame so the crop follows the speaker. Assign shot framing across wide, mid, and close sizes depending on comedic emphasis or pacing.

Step 3: Cut List Compilation in TSV

Generate a TSV text file representing the entire edit. Each row contains a start time, end time, shot size, and label ("the edit is text").

Step 4: Simulator Screen and Asset Insertion

Instruct the AI to find branding graphics or open the simulator, capture app screenshots, and place them across exact frame ranges matching verbal cues in the cut list.

Step 5: Spatial Graphic Placement

Measure the speaker's right edge from the rendered footage to dynamically position graphic cards into the empty physical space of the background.

Step 6: Synchronized Word-by-Word Captions

Generate high-contrast captions (white text with black outline) displayed one word at a time, holding each word on screen until the next word arrives.

Step 7: Sync and Multi-Format Verification

Verify frame count at 30 FPS with audio and video 0 milliseconds apart. Re-run the text file to export 30-second, 60-second, and horizontal versions from the same raw tape.

When This Works (and When It Doesn't)

This pipeline works when you record single-take talking-head clips on your local computer and want to skip dragging clips across a timeline. As Lacey put it: “The file's in the computer, it doesn't upload anywhere. The only thing that's happening is the instructions are being pulled and it's doing all of the things. It's just running the computer.” Because the edit lives as plain text, re-exporting different aspect ratios takes one shell command.

The system breaks down when your video relies on multi-cam shoots, b-roll storytelling, or emotional grading that demands visual intuition. If your pacing depends on subtle non-verbal pauses rather than rapid spoken sentences, mechanical silence thresholding will slice off your deliberate beats.

What to Do With This

If you are a solo founder recording product demos or weekly updates, test this setup on your next raw recording:

First, install Whisper and FFmpeg locally via your terminal. Run a basic Python script that transcribes your raw file to extract exact word timestamps.

Second, generate a simple TSV file where each row marks your spoken sentences, stripping pauses longer than 300 milliseconds.

Third, write an FFmpeg command that reads those timestamps and slices the video into a 15-second vertical output. Claire Vo pointed out the speed of the output during the demo: “approximately 4 seconds later there was a really well-clipped vertical short-form video with captions, with jump cuts, with a little zoom when you made a funny face.” Start with tight cuts and synchronized captions before layering on automated face-tracking or simulator screenshots.