Build your own AI video studio

You are helping me set up a project that turns a topic into finished vertical explainer videos (1080×1920, for Reels, TikTok and Shorts), the way @youfoundbruno makes his. Work in the current folder. Explain each step briefly as you go, and show me what you'll install before you install it.

1. Ask me first

• The topic of my first episode, and my audience.
• My handle, and a look: dark & calm, light & editorial, or bold & minimal.
• Voice: do I have an ElevenLabs (or other TTS) API key? If not, use the OS's built-in voice for drafts (say on macOS, espeak on Linux).

2. Research before writing

• Use web search to collect the key facts and numbers for the topic. Keep a research/fact-sheet.md with each claim, its source URL, the date checked and a confidence level.
• Flag anything outdated or disputed, and never put an unsourced number in a script.

3. Script format

Write each episode as data (TypeScript or JSON): a list of scenes, each with a voiceover line plus visual fields. Scene types to support:

• hook: a big number that counts up in the first second, with a kicker and a label
• title
• stat: a number with context
• list: a checklist that appears item by item
• compare: bad vs. good
• timeline
• quiz: pause & guess, a 3-2-1 countdown, then the reveal
• code: a bad snippet that morphs into the fix
• bars: comparison bars
• flow: a diagram with moving packets
• cta: comment KEYWORD

Rules: hook in the first 2 seconds, about 60–80 seconds total, one idea per scene, and the fix always shown.

4. Build the renderer

• Use Remotion (React → MP4). One composition per episode, at 30 fps.
• Respect platform safe zones. Keep text out of the top ~230px, the bottom ~470px and the right ~150px, where the app's UI sits.
• Burned-in captions, 3–4 words at a time, with the active word highlighted. Most people watch muted.
• Scene length = voiceover length + a short pad. Generate the voice first, measure each clip with ffprobe, then size the scenes to fit.
• Add subtle sound effects (a whoosh between scenes, a hit when the big number lands, ticks on the countdown). Synthesize them with ffmpeg so there are no licensing issues.
• Put theme tokens (colors, fonts, radius) in one file, so the whole look can change in one place.

5. Scripts to add

• voiceover: TTS per scene with caching, so unchanged lines aren't regenerated (or re-billed)
• sheet: a contact sheet of one still per scene, for a quick visual review
• render: every episode to out/

6. First run

Make one episode end to end, render a contact sheet, review it with me, fix what looks off, then render the MP4.

Rules

• Ask before installing anything or using paid APIs, and give me a cost estimate for TTS.
• Keep the code small and readable. No editing timelines; everything is code.
• If the topic touches law, health or money, add a visible "education only" disclaimer.