MICHAEL
0 of 8 steps done
AI Q · Guide 7

Thought to motion

I give Claude one thing. It comes back as an edited video in my brand, and nothing renders until I say yes.

One thoughta video take, a voice note or a few lines of text
Claude plans itfinds the beats
Gate 1you approve the plan
Gate 2you approve the stills
Your videoin your brand, with captions
Time an afternoon to set up, then a few approval rounds per video Cost the skill is free and open source; Claude Code is on paid Claude plans Level the most technical guide in the series

Prologue

The thought is the product

Guide 6 closed the loop around each post. This last guide is about the post itself, when it's a video.

  1. 1Set up Claude
  2. 2Give Claude a job
  3. 3Define your brand
  4. 4A content inbox
  5. 5A free publishing stack
  6. 6Timing and comments
  7. 7Thought to motion

Where you are

A brand, an inbox, a publishing stack and a feedback loop. Missing: one thought into a finished video, without an editor.

What you'll have

  • The thought-to-motion skill, set up
  • Your brand in one small file
  • A first video with no camera
  • Two approval gates you control
  • Captions in your language
  • A talking-head take that edits well

How I came to it

When I posted a video made with it, the most common question in the comments was how I do it. This guide is my answer.

Before you start

What your computer needs

Everything runs on your own computer. Your ticks are saved in this browser.

Not sure? Ask Claude Code: "Check if ffmpeg, Node 18+ and Python 3.10+ are installed on this computer, and tell me what's missing."

The lighter path

No camera needed to start. Text or a voice note becomes a video with captions, scenes and music.

The picture

One thought, two gates, one video

Tap a box to jump to its step. Boxes turn teal as you finish them.

A pipeline: your recording goes to prep, which writes a transcript with word timings; Claude finds the beats; at Gate 1 you approve the plan; the plan is validated; stills are made and checked by an independent reviewer; at Gate 2 you approve; the video renders, using your brand file, into an MP4 and a captions file Your recordingvideo, voice or text Preptranscript + word timings Claude finds the beatshook, contrast, reveal Gate 1you approve the plan Validaterules, safe area, face Stills + reviewerkey frames, fresh eyes Gate 2you approve the stills Renderdraft first, then final MP4 + captionsvideo + .srt file Your brandbrand.json · step 2

The coral boxes are where it stops and waits for you. On a narrow screen, swipe the picture sideways.

The setup

Eight steps

Steps 1 to 6A first video, no camera
Step 7Talking head
Step 8Optional
1

Get the skill

Open Claude Codein the folder for your videos
Run the two commandsor paste the prompt
Skill readyin .claude/skills

Or copy it from the repo.

npx skills add michaelburshtein/thought-to-motion-skill
cd .claude/skills/thought-to-motion && bash scripts/setup.sh
Install the thought-to-motion skill from github.com/michaelburshtein/thought-to-motion-skill into this project's .claude/skills folder, then run its setup script. If ffmpeg, Node 18+ or Python 3.10+ is missing, stop and tell me what to install.

The skill is MIT licensed. It renders with Remotion, which larger companies need a licence for. Read Remotion's licence before business use.

How I do it

I built it with Claude. Everything runs on my computer.

  • No uploads
  • No cloud transcription
  • No AI avatars
  • No voice cloning
Screenshot coming: Claude Code after the setup script finishes
Task
2

Put your brand in one file

brand.example.jsonbrand.json
Wordmark
[your name or brand]
Series label
[a short label]
Handle
[@yourhandle]
Brand line
[your brand line]
Theme
warm / cinematic / tech
Fill it from your guide 3 brand doc, or let Claude do it.
Fill in brand.json for thought-to-motion from my brand doc: [attach or paste your brand doc].
Wordmark: [your name or brand]. Series label: [a short label for your videos]. Handle: [@yourhandle]. Brand line: [your brand line]. Theme: [warm/cinematic/tech].
Show me the file before you save it.
  1. Key lineits own opener
  2. Brand introshort
  3. Your videothe thought
  4. Outrohandle + brand line
Every video with a brand file. Without one: no header, intro or outro.
How I do it

My outro always says "Life has its own soundtrack". My opener is a line I actually say in that video.

Task
3

Make your first video from text or a voice note

Pick a thoughtone you would say in under a minute
Type it or say ita voice note goes to your computer
Paste a promptbelow, and fill in the brackets

Your words are never rewritten, only split and timed.

Turn this text into a Reel with thought-to-motion: [paste your text]. Theme [warm/cinematic/tech]. Captions in [language]. Music [calm/warm/pulse/off].
Turn this voice note into a Reel with thought-to-motion: [file]. Theme [warm/cinematic/tech]. Captions in [language]. Music [calm/warm/pulse/off].

The skill's music is copyright-free. For a real song, set music to off and add it when you post (guide 5).

How I do it

In Hebrew, my captions are bilingual: Hebrew on top, smaller English under it. Hebrew uses the ivrit.ai model. Other languages Whisper knows work too.

Task
4

Read Gate 1: the plan table

Before anything renders, Claude shows you a plan like this. Read every row, fix what's wrong, then say yes.

Gate 1 · The planWaiting for your yes
Example plan table with placeholder content
TimeWhat is saidWhat appearsWhy
0:04"A word you stress"Keyword card with that wordNames what you're saying
0:09"This one, or that one"A/B compare, side by sideShows the contrast
0:21"Your last line"End card: handle + brand lineCloses in your brand
Also shownEvery caption fixProposed opener lineAny fact to add, with a source
Your yesOn screen often enoughCaptions readable in timeClear of app buttonsOff your face
Placeholder rows. After your yes, a validator checks these rules.
Gate 1 notes:
- Row [number]: [what to change].
- Caption fix: "[what it wrote]" should be "[what I said]".
- Opener line: [the line I want to open on].
Everything else is approved. Go on to the stills.
How I do it

A good plan is dense: something on screen every 2 to 5 seconds. If the table looks thin, I ask for more.

Task
5

Check Gate 2: stills and a second opinion

Stills sheet · placeholder STILL 1STILL 2STILL 3STILL 4
What the apps coverYour face
  • Text in the safe area
  • Nothing over your face
  • Captions in order
  • Fonts loaded
Stills sheetkey moments on one sheet
A separate reviewerblind to Claude’s reasoning
Claude fixesthe reviewer’s blockers
Gate 2you say yes, or ask for changes
Gate 2 notes:
- Still [number]: [what's wrong, for example "the card covers my mouth"].
- [Anything the reviewer flagged that I disagree with, and why].
Fix these and show me the sheet again. Don't render until I approve.
How I do it

The reviewer never approves for me. Its verdict is one more thing I read before I decide.

Screenshot coming: a stills sheet with the reviewer's verdict
Task
6

Render a draft, then the final

Quick drafthalf resolution. Watch start to end for lip sync and clipped words.
Your yesthen ask for the final
The finalfull video, a posting copy with measured loudness, .srt captions per language
The draft is approved. Render the final in [9:16 / 16:9 / both]. Tell me the measured loudness of the copy for posting, and where the video and caption files are.
How I do it

The final renders once, after my yes. Then it goes through my approval flow like any other post.

Task
7

Film a talking-head take

Key line firsta short pause before and after it. It becomes your opener.
Tripod, eye levelone fixed shot is enough
Say it oncenaturally. Long pauses are cut for you.
File to computerthen paste the prompt below
Edit this talking-head video with thought-to-motion: [file]. Opener line: [the key line I said]. Captions in [language][, with [second language] under it]. Format [9:16 / 16:9 / both]. Music [calm/warm/pulse/off].
WIDE MEDIUM CLOSE
  • Wide: the whole shot
  • Medium: a crop
  • Close: on your face
Virtual cameras, cut on your pauses. Crops of your real footage, not AI-generated angles.
How I do it

One take, one position, no cutaway shots. The skill adds the cards, the reveals and the camera changes.

Task
8

Optional: replace the background

Lock ittripod, exposure and focus
Clean platea few seconds of the empty room, then walk in
Your own photosame place, light and camera height as your shot
Preview firstone or two frames
Replace the background in [file] with [my photo of a room]. The take starts with a clean plate of the empty room. Show me a preview of 1 to 2 frames first, then run the quality check after processing and show me the worst frames.

A plain wall behind you gives the cleanest result.

How I do it

My best results came with a clean plate.

Task

Check it works

Watch your first final like a viewer

Watch the final on your phone with the sound on, once. Then tick these.

Task

If something is off, tell Claude what you saw and at what second. It fixes the plan and renders a new draft.

My mistakes

What went wrong for me

Each one is now a rule in the skill.

What happened: My voice drifted from my lips

Each cut lost a few milliseconds of sound. By the end, about half a second.

Cuts land on whole frames. A render won't deliver if sound and picture disagree.

What happened: I rolled back a whole upgrade

A batch of features, built in one day. Good on paper, but I preferred the old version.

Each upgrade renders side by side with the current version, on the same video. I decide, by ear too.

What happened: A thin test proved nothing

A test without the usual cards looked flat and said nothing.

I test only inside a full, real plan.

What happened: The opener clipped my last word

The transcript's word times ran a little early.

The opener ends in the pause after the word, checked on the audio.

Stay safe

My do's and don'ts

Do

  • Start with text or a voice note before you film.
  • Review drafts, and render the final only once.
  • Use your own photos, or images you have the rights to.

Don't

  • Skip a gate to save time, or let a fact you didn't check reach the screen.
  • Embed songs, meme sounds, or characters and logos that aren't yours.
  • Post a final you haven't watched with the sound on.

Epilogue

The end of the road

One thought went in. A video came out, in your brand, with captions in your language, and you approved every part of it on the way.

Your inboxguide 4
Your publishing stacksends it out, guide 5
Your looptells you what to make next, guide 6

This series started with one comment under a video made with this skill. Someone asked how I use it, and I said I'd put out a short, simple guide. Seven guides later, you have the whole road.

The series continues soon. I share what I learn as I build it.

Built on the work of others: Remotion for rendering, DeepFilterNet for clean speech, ivrit.ai for Hebrew transcription, OpenAI Whisper for other languages, and MODNet for background replacement. The skill itself is on GitHub, MIT licensed.