The argument in one line.
A single PDF of 106 locked Remotion style rules lets two prompts turn Claude Code into a full video editor — generating cuts, captions, and transparent motion graphics from one talking-head recording without a traditional NLE.
Read if. Skip if.
- A solo creator who records talking-head videos and wants Claude Code to fully replace DaVinci Resolve, Premiere, or CapCut for motion-graphics editing.
- Someone already comfortable running Claude Code sessions who wants a reusable, PDF-driven style system instead of prompting a new look from scratch every time.
- A YouTuber or short-form creator who wants automated animated captions and transparent overlay graphics generated from the same transcript used for scene cuts.
- You haven't installed Claude Desktop, Git Bash, or Node.js yet and aren't willing to do basic environment setup before this workflow makes sense.
- You want a no-code drag-and-drop editor — this is a prompt-driven, code-generated pipeline, not a GUI timeline.
The full version, fast.
A free downloadable PDF acts as a 'vault': one prompt makes Claude Code absorb 106 locked Remotion editing styles, transcription rules, and revision instructions in a single pass, so nothing has to be re-explained per session. A second prompt supplies the raw talking-head video plus one style's activation phrase, and Claude uses Whisper.cpp to time-align the transcript, decide scene cuts, ask whether the format is vertical or horizontal, then renders the full edit — talking head, generated motion graphics, animated karaoke captions, even transparent-background overlays — with no DaVinci Resolve, Premiere, or CapCut involved. Revisions are just more prompts against the same project, and results improve once you protect good versions, get specific about placement, and switch to the strongest model when troubleshooting.
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.

01 · 106 Claude Remotion Styles
Cold open: the pitch that Claude + Remotion can be taught entire editorial styles, not just individual motion graphics.

02 · What This Remotion Tutorial Covers
Overview of the video: a vault of 100+ styles the viewer can scroll and pick from, plus a note to watch the chapters.

03 · Changed My Life
Creator thanks viewers/supporters of the previous Claude+Remotion video before continuing.

04 · Step 1: Claude Setup
The four prerequisites: Claude desktop app, Git Bash, Node.js, and the downloadable project folder; pointers to short fix-it videos if any step fails.

05 · What To Expect Next
Frames the rest of the video: not just installing Remotion, but giving Claude a full style vault and eliminating manual editing entirely.

06 · Step 2: Understanding Claude Vault
Walkthrough of the vault PDF: install buttons, viewer instructions, hidden Claude-only instructions, 106+ styles with activation phrases, and a separate revision-prompt document.

07 · Step 3: Start Your Claude Project
Set the project folder in a new Claude Code session and switch permission mode to auto so the run doesn't stall on repeated approval prompts.

08 · Step 4: Activate Claude Remotion Vault
The first prompt — 'read the PDF, activate chron_vault' — and Claude's confirmation summary of everything it absorbed.

09 · Step 5: Plan + Build First Edit
Second prompt: pick a style, paste its activation phrase and the folder path, answer Claude's horizontal/vertical question, say go, and review the generated scenes before completing the edit.

10 · Step 6: Add A Split-Screen With Remotion
Revision prompt turns the alternating talking-head/graphics cut into a permanent split-screen layout.

11 · Step 7: Switching Remotion Styles
Generating fresh variants of the same base edit by swapping the activation phrase while protecting the liked version.

12 · Step 8: Create Transparent Background
Generating transparent/greenscreen-style talking-head overlays inside Claude, replacing the DaVinci Resolve background-removal step from the creator's prior video.

13 · Step 9: Add Captions In Remotion
Same Whisper.cpp transcription reused to generate animated, word-synced captions positioned with a plain-English instruction.

14 · Step 10: Add Images To Your Edit
Dropping custom images into the project folder and having Claude incorporate them into a new version of the edit.

15 · Step 11: Fix Common Issues (Example)
Live troubleshooting example: generated graphics were crowding out the talking-head shot, fixed with one descriptive prompt on the strongest model.

16 · Step 12: Five Secret Claude Tips
Five troubleshooting rules: be descriptive, don't fight a bad session, protect good versions, be specific with revisions, use the strongest model.

17 · Special Thanks To Supporters
Sign-off: every asset (PDF, prompts, revision prompts, setup links) is free, thanks to supporters, and a like/comment/download CTA.
Lines worth screenshotting.
- A single PDF handed to Claude Code can front-load 106 locked editing styles plus revision rules in one prompt, so style logic never has to be re-explained mid-session.
- Claude uses Whisper.cpp to time-align a talking-head video's transcript, then decides where scene cuts happen and what each generated visual represents from that timing.
- Setting Claude Code's permission mode to auto removes the repeated 'allow / allow / allow' prompts that otherwise interrupt a long generation run.
- One activation phrase, like 'CHRON_STYLE_27', is enough to lock a specific visual style before Claude starts generating scenes.
- Explicitly telling Claude to leave a liked version untouched before generating a new one is what prevents good output from being silently overwritten.
- Transparent-background, greenscreen-style motion graphics can be generated directly inside Claude Code, removing the need for a separate compositing tool.
- Animated, word-synced captions are generated from the same Whisper.cpp transcription already used for scene timing — one transcript, two deliverables.
- Custom images can be woven into a generated edit just by naming the file and folder path in a prompt, with no manual placement.
- When a Claude session gives bad results from the very start, restarting with a clean session outperforms continuing to fight the same confused context.
- Vague revision requests like 'make it better' get worse results than specific, spatial instructions like 'move the head to the right, keep captions bottom-middle.'
- Switching to the strongest available model for troubleshooting is worth the cost — saving credits with a weaker model usually means fixing the same problem twice.
- A common Claude+Remotion failure mode is generated scenes crowding out the talking-head shot almost entirely; naming that imbalance directly is enough for Claude to rebalance the layout.
A prompt-driven replacement for your video editor
Front-loading style rules into one PDF, then driving revisions with specific spatial prompts and a strongest-model fallback, turns Claude Code into a full multi-style video editor.
- The whole workflow only requires four installs: Claude desktop app, Git Bash (so Claude Code can run local Git sessions), Node.js (for Remotion), and one shared project folder/PDF.
- When Claude reports a missing dependency it already has installed, treat it as a known quirk — short, targeted install videos fix these environment issues faster than debugging blind.
- The vault PDF bundles four things Claude needs in one document: install links, plain-English usage steps, hidden instructions meant only for Claude, and 106+ style options with activation phrases.
- Style descriptions in a reference document are worth reading, not just skimming the thumbnail — the image is only a rough representation of what the style can actually produce.
- A separate revision-prompt document is worth preparing in advance specifically for troubleshooting, so fixes don't require improvising a prompt from scratch when something looks off.
- Point an agent at a dedicated project folder before doing anything else, and set the permission mode to auto so a long multi-step generation doesn't stall on repeated 'allow' prompts.
- A single prompt — 'read the PDF in this folder, then activate [system name]' — is enough to make Claude absorb dozens of rules, skills, and versioning logic at once.
- Confirm the load actually worked by having Claude summarize what it absorbed; a response that's noticeably thinner than expected is a signal to re-prompt before continuing.
- Choosing a preset is just pasting its activation phrase into the next prompt along with the project folder path and the raw input file.
- An agent can use a transcript's exact timing to decide where cuts happen and what each generated visual represents — that's interpretation, not verbatim captioning.
- An agent will ask a clarifying question before building a deliverable — answering it correctly the first time avoids a wasted full render.
- A layout change to an existing output is a single descriptive revision prompt, not a redo of the whole project.
- Reusing a stock prompt template for a common revision type made the change consistent and fast to specify.
- New variants of an existing output are generated by explicitly telling the agent to keep the current version untouched and create a new one with a different setting.
- Re-rendering the same base project in multiple styles just by swapping one parameter proves the style and the underlying structure are decoupled.
- Complex compositing (transparent/greenscreen-style overlays) can be generated directly by an AI agent, eliminating a dedicated tool for that step.
- Results aren't guaranteed on the first prompt — sometimes it takes one prompt, sometimes two or three; that variance is normal and not a sign the setup is broken.
- When an agent only completes part of a multi-part request, a direct follow-up prompt naming exactly what's missing closes the gap.
- The same transcription used for cut timing can double as the source for animated, synced captions — one transcript, two deliverables.
- Caption placement and animation style can be controlled with plain-English instructions, not manual keyframing.
- Custom assets dropped into a project folder can be woven into a generated output just by naming the file and folder path in the prompt — no manual placement required.
- A common failure mode — generated elements crowding out the primary subject — is fixed by directly naming the imbalance and describing the intended result.
- Descriptive troubleshooting prompts (what's wrong + what you want instead) consistently outperform vague ones ('make it better') for getting a usable fix in one pass.
- Be descriptive: state what feels wrong and what you want the end result to feel like, not just 'make it better, make it better.'
- If a session starts off bad, don't fight it — start a fresh session rather than trying to recover a context that's already confused.
- Always protect versions you like by telling the agent to leave them untouched before generating a new variant, so experimentation can't silently overwrite your best result.
- Be extremely specific with revisions — name exact positions and even percentages instead of general direction.
- Use the strongest available model for troubleshooting; saving credits with a weaker model costs more time when the same problem has to be fixed twice.
Terms worth knowing.
- Remotion
- A React-based framework for programmatically generating video, letting code — including AI-written code — define motion graphics, transitions, and captions instead of a manual timeline editor.
- Claude Code
- Anthropic's command-line AI coding agent, used here not to write app code but to generate and render an entire video-editing project end to end.
- Whisper.cpp
- A lightweight, offline speech-to-text engine that Claude runs locally to transcribe a video with word-level timestamps, powering both scene-cut timing and captions.
- Style vault
- A single reference PDF containing dozens of pre-defined visual style templates and rules that an AI agent is taught once and can then recall by name.
- Activation phrase
- A short code, like 'CHRON_STYLE_27', pasted into a prompt to tell Claude which pre-defined style from the vault to apply to the current render.
- Auto mode
- A Claude Code permission setting that lets the agent execute a full multi-step task without pausing to ask for approval on every individual action.
Things they pointed at.
Lines you could clip.
“The real power is when you teach Claude style.”
“You need Claude desktop app... Git Bash... Node JS... and finally, you just need the one folder that I'm giving you for free.”
“The more descriptive I am with identifying the issue and then suggesting the fix, the better result I will get.”
“Don't waste time on the weaker models trying to save credits. Just fix it once and for all so you can move forward.”
“The biggest thing is don't treat Claude and Remotion like a magic button right at the beginning.”
Word for word.
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The bait, then the rug-pull.
The pitch is blunt: swap DaVinci Resolve, Premiere, and CapCut for a single PDF that teaches Claude Code 106 locked Remotion editing styles, then let it cut, caption, and composite a finished video from a talking-head recording using little more than two prompts.
Named ideas worth stealing.
The Two-Prompt Vault Workflow
- Prompt 1: 'read the PDF in this folder, then activate [vault name]' — Claude absorbs 106 styles, composition/versioning rules, and skills in one pass.
- Prompt 2: paste the project folder path, the raw video/script file, and one CHRON_STYLE_X activation phrase, then answer Claude's orientation question.
- Claude decides scene cuts and visual concepts from the transcript's exact timing, not verbatim captioning.
- Say 'go' — Claude builds and renders the full edit; 'complete this edit' merges the talking head with the generated scenes.
The entire workflow compresses to two prompts once the vault PDF exists: one to teach Claude the whole style system, one to build a specific video.
Five Claude + Remotion Troubleshooting Rules
- Be descriptive — say what feels wrong and what you want, not just 'make it better.'
- If a session starts off bad, don't fight it — start a new session for a clean slate.
- Always protect versions you like — tell Claude to leave them untouched before generating a new one.
- Be extremely specific with revisions — name exact positions/percentages, not vague direction.
- Use your strongest model when troubleshooting — don't save credits with a weaker model and fix things twice.
A short, reusable checklist for getting reliable output from an AI agent doing generative creative work, not just code.
How they asked for the click.
“$0. All of it. The style vault, vault prompts, revision prompts, and setup links — everything linked below.”
Explicitly frames every asset as free to remove purchase friction, then stacks a like/comment/download ask at the very end (25:55) alongside a repeated 'everything linked below' pointer. Download is a Gumroad link; a separate PayPal donate link is offered as optional support rather than a paywall.































































