The four frames that stop the scroll.

Familiar × impossible × instant payoff, delivered in four frames, drawn from three public reels with 1.7B+ combined plays. Score your own hook.

An AI video hook that stops the scroll runs in four frames (familiar, fracture, reveal, payoff), and the reveal has to land within the first 2 to 3 seconds.

Familiar × impossible × instant payoff

Take three public reels we studied, with play counts snapshotted on 22 Sep 2026: @cat_vlog365 at 495,021,428, @bennettwaisbren at 1,019,816,251, and @stat.us.ai at 213,154,735. Combined: 1,727,992,414 plays. MakeRoll hosts no copy or still of these reels; they are attributed study references, and their creators have not endorsed anything here.

Strip the subjects away and the same skeleton is under all three. We call the pattern familiar × impossible × instant payoff, delivered in four frames. Median AI TikTok: 418 views (Riffkit 2026). These three reels are the tail, not the norm. The pattern explains the tail.

Each frame has one job.

Fail one and the swipe happens there.

Frame 1, Familiar

Something the viewer already trusts: a face, a room, a format they scroll past daily. The vlog framing, the mirror selfie, the talking head. Familiarity buys the first half second. It costs nothing and AI renders it convincingly now.

Frame 2 · Fracture

The impossible enters the familiar. The vlogger's kitchen is now a cliff face. The mirror shows a different person. One element breaks the viewer's model of the scene while everything else stays plausible. The actual hook is a contradiction the brain must resolve, and resolving it requires watching.

Frame 3 · Reveal

Pay off the contradiction fast, within the first 2 to 3 seconds. Show what the fracture really is, or escalate it once. The reveal is where AI video genuinely outperforms live action: the impossible thing is not a stunt, it is a generated frame that would cost thousands practical.

Frame 4 · Payoff

Give the viewer a reason to finish: the transformation completes, the joke lands, the loop closes back into frame 1. Loops are the multiplier. A reel that ends where it starts gets re-watched, and platforms count re-watches as watch time.

Variation is cheap. Taste is the bottleneck.

Live-action crews spend budget on the fracture: rigs, sets, VFX. AI pipelines spend prompts on it. That inverts the economics: the familiar frame is nearly free, so all your craft goes into one impossible element and its reveal.

Our documented pilot produced 3 concepts and 6 exports from metered calls, QA scored 91 to 96 (pilot ledger). Metered calls cost little per attempt, so you can afford to generate five fractures and critique-pick one. That is the whole advantage: variation is cheap, taste is the bottleneck.

Where this breaks

• Familiar too long. If frame 1 runs past a second, there is no hook yet. The viewer scrolled for a reason.
• Fracture with no reveal. A contradiction that never resolves reads as a glitch, not a hook. The reveal is how you tell the viewer you meant it.
• Impossible without familiar. Surreal with no anchor is wallpaper. The pattern is a product: familiar times impossible, not impossible alone.
• Synthetic realism, undisclosed. When the familiar frame is a realistic AI person, label it. Our disclosure rule: an on-screen AI-character card in the opening seconds and the caption stating it (full checklist). Every character in our own demos is an original fictional adult AI character created for the MakeRoll demo.

Critique makes taste visible

A pattern on a page does not make anyone good at this. Running it weekly does: one brief, generate the four frames, get critique on which frame failed, ship the fixed roll. That loop is the weekly club. Membership is open now (see the plans).