The six sessions 🎬 · one artifact, built across all of them
Bring sixty seconds of your own footage. Something real that you would publish. By session 6 you have a video that holds attention, carries usable captions, and says how it was made.
The first three seconds
Where video actually loses people, and the one lever that costs nothing. Run a real silence detector, then watch both obvious improvements go backwards.
💸 Session 2 · easyWhat the tools cost
Per-second pricing, clip lengths and resolution caps for Veo, Runway, ElevenLabs and Descript. Plus the model that is being switched off next month.
🔒 Session 3 · mediumThe constraints they hide
English-only. Single-face-only. Two seconds maximum. Every real limit in this category is three clicks deeper than the page that sold it to you.
💬 Session 4 · mediumCaptions are not optional
The Level A escape hatch that disappears at Level AA, what auto-captions actually score, and the error-rate gap that holds across every vendor.
🧪 Session 5 · hands-onHow it fails, measured
Best model scores 29.5 out of 100 on physics. Identity and wardrobe drift are formally scored defects. And the iteration cost nobody has actually audited.
🏷️ Session 6 · hands-onSay it was made with AI
Article 50 has been law since August. What C2PA proves, what its own documents say it cannot, and what each platform actually requires of you.
Your passport 🎫
Clear all three quiz questions in a session and it gets stamped. Stamps live in this browser only.
Choose your path 🗺️
Three doors into the same conviction: the tools are perishable and attention is not.
What is real here 🧾
This course has a simulator, so it owes you a precise answer about which numbers are measured and which are modelled.
| Claim | Status |
|---|---|
| The silence detection, cut points and resulting runtime | Real. Computed from the audio envelope in your browser at the moment you click, not stored |
| The retention curve and its ladder | A teaching model, labelled as one on the widget itself. Directionally honest, not a measurement of any audience |
| Every tool price, clip length and resolution limit | Quoted from the vendor's own pricing or documentation page, each with the URL, dated August 2026 |
| Caption error rates and the disparities across speakers | Peer-reviewed. 221 hours across 11 services for the error rates; PNAS 2020 for the disparities, replicated on identical phrases |
| Physics and consistency scores | Published benchmarks, Physics-IQ and VBench-2.0, with the caveat that vendor documentation currently admits less than the literature measures |
| The iteration cost per usable shot | Deliberately not stated. No independent audit exists, and the one large public figure has an undefined denominator |
The knowledge map 🧠
The whole course at a glance - hover a session to spotlight its concepts, click any node to jump in.