What the "AI" in the name actually means
People search "caught at the window ai" because the versions flooding TikTok are obviously not filmed: a famous face sits in a window that never existed, eating a burger in a flat that was never rented. This page explains that AI version plainly: what generates it, what your photo contributes, and where the seams show.
Three ways the scene gets made
The trend exists in three production styles. Filmed edits restage the scene with a real camera and a cooperative actor. CapCut templates wrap the framing around a clip you supply, inside the CapCut app. The AI version (this site) generates the entire scene from a single still photo: no camera, no actor's afternoon, no editing timeline.
The AI route is the one that lets the subject be someone who was never filmed: a friend across the world, a parent who would laugh but never pose, or yourself on a day you stayed in. The generator page is where that happens.
How the generation works
Two inputs meet in the model. The motion reference is a short clip of the scene performed by a stand-in; it defines the push-in camera path, the beat timings and the room. The photo defines the person. A fixed prompt (the prompt guide walks through it) instructs the model to follow the reference for movement only, and to rebuild the person from the photo in every frame.
The output is a 12-second, 720p, vertical video. Generation takes a few minutes of model time per video, which is also why a render costs credits; the free page shows the math.
What to expect, honestly
A clear photo yields a recognisable star and the stare lands. Busy photos, sunglasses, or a face in profile give the model less to hold onto, and likeness drifts. Small artifacts (a wobbly hand, a background object that changes between frames) are the current price of the technology and read as part of the format's charm at phone size.
The result is a fictional scene, and it should be posted as one: a joke among consenting adults, not evidence anyone was anywhere. Cast your star in the generator, and read the trend page for how the format took off.