The gag, explained

The Invisible Moped

Look at any clip from the Crazy Frog AI trend and one detail does the heavy lifting: the rider is sitting on nothing. Legs bent, hands gripping unseen handlebars, body leaning into turns — and beneath them, empty air. Every so often a lone wheel floats past, or a headlight blinks into existence for a frame, then thinks better of it.

That is the invisible moped. It is the difference between "AI made an action video" and a joke you can recognize from across the room.

Why the bike is invisible (and the rider isn't)

It is not a stylistic choice anyone made — it is how the physics of the prompt works out. The videos are generated from a single reference photo of a person. The model is told, in effect: this exact person, squatting low as if riding a moped, at speed, chased by a cop. Two things follow.

• The person is anchored. The reference photo pins the face, the hair, the outfit. So the rider renders sharp and consistent — that is the whole point of the reference.
• The bike is not anchored. No photo of a moped is provided, and "as if riding" is weaker conditioning than "this person". The model spends its coherence budget on the rider and lets the machine under them stay vague — ghosted handlebars, a wheel that surfaces when the camera angle happens to demand one.

The result reads as a deliberate deadpan gag: maximum-effort chase physics, zero vehicle. Nobody designed the joke; the joke fell out of the tool. Then everyone kept it, because it is funnier than a visible bike would be.

Where it sits in the six-shot storyboard

The chase follows a fixed structure — cop close-up, squat ride, look back, backflip, getaway, tiny dot — and the invisible moped plays a different role in each. It is funniest in the squat ride (full body, full speed, no bike) and the backflip (there is nothing to flip off of, and yet). The getaway is where the floating wheel usually makes its cameo. By the final shot the rider is a dot on an empty road, and the gag resolves: the cop lost, the bike never existed, your face did the whole stunt.

Make your own

The DIY route works in any general AI video model — write the chase prompt, keep the "as if riding, moped invisible" line in it, and the model will do the rest. The tutorial has the full copy-paste prompt. Or take the one-click route: upload one photo here, pick one of 17 backgrounds, and download the finished 14-second clip.

Tip: front-facing photos with the whole upper body in frame give the best squat — the model has more of you to work with.