At a glance
WHAT IT’S REALLY ABOUT
Claude Code turns raw takes into TikTok edits, hands-free
- Brooke Lacey demonstrates a Claude Code workflow that edits short-form videos locally—using tools like FFmpeg, Whisper, and Python—without a traditional timeline editor.
- Her editing pipeline optimizes for TikTok-specific humor and cadence by combining silence detection with word-level timing to place jump cuts precisely.
- The system treats “the edit as text” via a TSV cut list, enabling quick rerenders into multiple lengths and aspect ratios from the same raw take.
- Brooke trains Claude on her style using her existing TikTok corpus and comment threads to both improve edits and generate new script ideas.
- When Claude fails, she uses an OS-like structure of instructions and status files (“lane OS”) to debug, re-align behavior, and propagate fixes globally.
IDEAS WORTH REMEMBERING
5 ideasClaude Code can function like a hands-free, terminal-driven video editor using existing local tools.
Brooke uses Claude Code as an “orchestrator” that calls local tooling (FFmpeg, Whisper, Python) to detect speech/silence, align word timings, track faces for smart crops, and render multiple aspect ratios—without a traditional NLE timeline.
Great short-form editing is a timing problem—word-level alignment matters as much as silence detection.
For TikTok-style humor, the exact cut boundary often depends on precise word-end timing; Brooke combines silence detection with Whisper word timestamps to land jump cuts where cadence and facial expression “hit.”
Representing edits as text unlocks fast iteration and multi-format rendering from one recording.
Her system outputs a simple TSV/text cut list (start/end/shot size/labels). Because the “edit is text,” she can rerun the same command to produce 15s/30s/60s and landscape variants from the same source take.
Once timing and composition are structured, AI can reliably automate polish: overlays, captions, and zooms.
Claude inserts app screenshots and on-screen “cards” based on the cut list—measuring frame composition (e.g., empty wall space) so overlays fit naturally next to her, plus one-word-at-a-time captions with consistent styling.
A creator’s best content ideation dataset is their own posts plus comments—Claude can mine both with the right access pattern.
Because TikTok is hard to scrape at scale, Brooke builds a workaround: she downloads her own TikToks to Google Drive for analysis, then has Claude read comment threads while logged in to generate new script ideas aligned to what viewers ask.
WORDS WORTH SAVING
5 quotesThe 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.
— Brooke Lacey
And not only that, but it's listening and goes, "Oh, I just heard you say cut that out." And then it also was like, "You repeated this a couple of times, and I took the one that was the best cadence." And I was like, "What?"
— Brooke Lacey
This is probably the 15th or 16th time in the last six months that I literally, like, fell to the floor. Like, I'm so dramatic anyway, but I was like, "What just happened?"
— Brooke Lacey
No timeline, no project file, no editor, and because the whole edit is a text file, the same command makes the 30 and 60 and the landscape versions from the same take.
— Brooke Lacey
In the first part of the video, I say, "I've been in tech for 25 years, and I've never been more scared for people." And that little laugh right there, where I'm like, "Scared for people," I had to tell Claude, "Did you hear the cadence in my voice?" Because that is how a hook works for me.
— Brooke Lacey
High quality AI-generated summary created from speaker-labeled transcript.
