CHAPTERS
- 0:00 – 1:16
Claude Code edits short-form video with “no hands”: why this is a big deal
Brooke and Claire open with the headline claim: Claude Code can take a raw take and produce a polished vertical short with TikTok pacing—without a traditional NLE timeline. They tease the core value: precise jump cuts, cadence-aware selections, and on-device processing that feels like magic to creators who are tired of manual editing.
- •“No timeline, no project file, no editor” framing for terminal-driven editing
- •Claude reacts to spoken cues like “cut that out” and removes repeats automatically
- •Outputs feel native to TikTok: captions, jump cuts, zooms, pacing
- •Emphasis on automation that still preserves creator style/vibe
- 1:16 – 3:37
Show setup: why Brooke built a Claude-based editing stack for TikTok-quality output
Claire introduces Brooke and the goal: a creator-focused Claude Code workflow that produces high-quality short form (not generic AI content). They position the conversation around solving the grind of short-form editing and scaling a creator “empire-building stack.”
- •Brooke’s TikTok scale (200K+ followers) and creator needs
- •Focus on humor, timing, and jump cuts as the hard part to automate
- •Using Claude Code for both editing and ideation/scripting
- •Motivation: reduce time cost while keeping quality high
- 3:37 – 4:59
Brooke’s background: why short-form jump cuts are uniquely hard to outsource
Brooke explains her long history with video and why short-form editing is so tedious: comedy depends on exact timing, facial expressions, and micro-cadence. After struggling to delegate to editors, she kept editing herself—until Claude made the precision scalable.
- •Longtime video experience (news/early social video era)
- •TikTok comedy relies on “cut at the exact moment” jump-cut timing
- •Editors struggled to match her natural cadence and expressions
- •Manual editing took hours even for simple talking-head videos
- 4:59 – 9:24
The first “Claude cut my video” moment: a 25–30 minute take becomes usable
Brooke recounts recording a long, messy take and verbally noting mistakes (“cut that out”) as she went. After AirDropping the file to her Mac, she tried Claude Code—and it not only cut mistakes, but chose the best repeated phrasing by cadence.
- •Recording long takes with spoken self-direction
- •AirDrop workflow to get phone video onto Mac quickly
- •Claude detects “cut that out” moments and removes them
- •Claude selects best takes among repeats based on cadence
- 9:24 – 11:06
Live demo: triggering the terminal workflow and seeing a finished vertical short
They run the workflow live: a raw video is pulled up and within seconds a clipped vertical ad appears with captions, jump cuts, and comedic zooms. Claire narrates what listeners can’t see and asks how Brooke “pre-populated” Claude to do this reliably.
- •Terminal-first workflow for maximum system access
- •A single command/prompt initiates the full edit pipeline
- •Near-immediate generation of vertical short with captions and zooms
- •Discussion of whether this is a skill, prompt, or repeatable template
- 11:06 – 13:09
Inside the edit chain: silence detection + Whisper timing + face tracking + cut lists
Brooke walks through the pipeline Claude describes: detect speech segments, refine cut boundaries with Whisper word timing, track her face for dynamic cropping, and generate a text-based cut list (TSV). The result is an edit that can be re-rendered into multiple formats from the same source.
- •Speech/silence detection to find talk segments quickly
- •Whisper word timing for clean cut edges at word endings
- •Face tracking per frame to keep the vertical crop locked on subject
- •TSV cut list as the “edit is text” source of truth
- 13:09 – 15:59
Graphics, app screens, and captions: turning a take into an ad-ready asset
The workflow goes beyond cuts: Claude inserts app screenshots at specific moments, places “cards”/graphics based on measured negative space, and generates one-word-at-a-time captions with styling. Brooke highlights the on-device nature of the toolchain (FFmpeg, Whisper, Python) and multi-version exports.
- •Automatic insertion of app screens/simulator screenshots into the edit
- •Layout-aware graphic placement using measured wall/empty space
- •Word-by-word captions with white text + black outline styling
- •Multi-aspect and multi-length renders (15/30/60, landscape) from the same cut list
- 15:59 – 18:01
Why most “make it funny” prompts fail: encoding short-form structure and pacing
Claire contrasts her vague prompting with Brooke’s structured definition of what makes a good TikTok edit. They call out specific creative rules—shot-size variation, exact cut timing, caption behavior—that Claude can execute when made explicit.
- •Short-form quality depends on structure, not generic humor instructions
- •Shot size variation (wide/mid/close) as a repeatable rule
- •Cut timing must align to audio/word endings to feel “natural”
- •Existing local tools can deliver pro results when orchestrated well
- 18:01 – 22:14
Building a TikTok corpus for Claude: downloading videos + pairing them with comment threads
Brooke explains how she collects her own TikTok videos into Google Drive for Claude to reference, since TikTok is hard to scrape directly. She pairs each video with live comment scraping (while logged in) so Claude can mine audience feedback and generate new content ideas.
- •TikTok scraping limitations; workaround via creator-owned downloads
- •Store videos in Google Drive so Claude Code can access them
- •Open TikTok to scrape comments from specific post IDs
- •Use comments as training signal for follow-ups and new scripts
- 22:14 – 23:41
Teaching Claude her hooks and cadence: turning personality into repeatable editing rules
They dig into how Brooke explicitly trains Claude to recognize her hook patterns, voice cadence, pauses, and comedic timing—down to a small laugh that signals a hook. The goal is not just better edits, but scripts and structures that match her on-camera persona.
- •Hooks are audible patterns (cadence, pauses, micro-laughs)
- •Claude learns what “sounds like Brooke” from repeated references
- •Style learning improves both cuts and future script generation
- •Limits remain (facial expressions are real), but structure is teachable
- 23:41 – 25:36
Two-character formats and “WTF Even Is”: scripting complex comedy structures
Brooke describes a recurring format where she plays two characters in dialogue to explain complex topics (e.g., blockchain) through absurdity and contrast. She notes it’s harder to train because the humor comes from tightly controlled back-and-forth pacing and escalating complexity.
- •Two-person (self-to-self) dialogue format for educational comedy
- •Humor depends on contrast: complex explanation vs confused reactions
- •Harder to script because absurdity must be precisely staged
- •Claude can help, but this format needs deeper structure training
- 25:36 – 28:18
Creator operations: posting cadence, live streaming plans, and ad experimentation
Claire recaps the system and asks about Brooke’s publishing rhythm. Brooke shares she posts about once a week due to a strong existing audience, but plans to increase live streaming and is currently producing ad creatives—prompting Claude differently for ad intent.
- •Once-a-week posting supported by mature audience and repost cycles
- •Plans to lean into live streaming for DevRel-style product walkthroughs
- •Using Claude-generated edits specifically for TikTok ad creative
- •Ad videos require different constraints and messaging than organic posts
- 28:18 – 32:52
Lightning round: Gen X perspective on tech shifts and why experienced builders should post
They discuss how Gen X lived through major platform shifts—from early PCs to the internet to Y2K—creating pattern recognition for today’s AI moment. Brooke encourages older technologists to share stories and knowledge on TikTok and similar platforms.
- •Generational “tech whiplash” experience informs AI adoption mindset
- •Anecdotes: Windows 3.1, early CS courses, internet explosion, Y2K
- •Call-to-action for experienced technologists to share publicly
- •AI is another major shift—jump in early and contribute context
- 32:52 – 34:57
When Claude disappoints: Brooke’s ‘lane OS,’ status files, and corrective prompting loops
Brooke explains how she handles failures using a structured “operating system” of instructions, memory lanes, and status markdown files. When Claude goes off track, she sends it back to the foundational instructions, diagnoses what it missed, and updates the global system so future sessions improve.
- •Custom “lane operating system” with memory lanes and write restrictions
- •Project-level status MD files + global instructions for consistency
- •Debug prompting: return to baseline, identify what was missed, retry
- •Propagating fixes to the global system so errors don’t repeat
- 34:57 – 36:33
Where to find Brooke and Zosha: closing and how to connect
Brooke shares where people can follow her and highlights her AI learning app, Zosha. Claire wraps with standard show outro and encourages comments, subscriptions, and listening on podcast platforms.
- •Brooke’s handle: BrookeJLacey across platforms (TikTok biggest)
- •App: Zosha (AI learning on mobile; free)
- •Open DMs and availability for meetups while traveling
- •Show wrap: subscribe, comment, and find the podcast feed
