A prompt you never have to write
Most AI video tools hand you a blank prompt box and wish you luck. Caught at the Window does the opposite: the prompt is part of the product, fixed on the server, tested against the trend's reference clip. This page explains what that prompt actually does, in plain language, so you know what you are buying when you press Generate.
If you would rather just make the video, the generator needs only a photo. If you want the scene itself explained visually, start with the template breakdown.
What the prompt tells the model
The fixed prompt has four jobs.
Treat the reference as a motion guide, nothing more. The reference clip supplies the camera path, the timing and the room: a night push-in toward one lit window, a small table, a takeaway burger meal. The prompt forbids copying any person from the reference: not the face, hair, skin tone, age or clothes.
Cast your photo as the person. The adult in your uploaded image is the one at the table. The prompt instructs the model to keep that face, hair, skin tone, body and outfit in every shot, which is why a clear, front-facing photo matters more than anything you could type.
Play the four beats in order. Glide in from the street. Sit at the table, burger in both hands, mid-bite. Notice the camera, freeze, and give it the annoyed stare. Set the burger down and pull the curtain shut, ending on the closed curtain with warm light at its edges.
Keep the frame clean. Photorealistic night lighting, warm indoors against the blue-dark exterior, one continuous shot, and a list of exclusions: no text, no brand logos or packaging marks, no subtitles, no watermark, no extra limbs. The exclusions sound dull; they are what stop a fun clip from turning into an uncanny one.
Why fixed beats free-form here
A trend is a shared staging. When every version of Caught at the Window opens on the same building and ends on the same curtain, viewers recognise the format in the first second and stay for the only variable: who got caught. Letting every user rewrite the scene would dissolve the very thing being searched for. The trend page traces how that staging spread.
Your leverage is the photo. Choose it with the best photo guide, then let the prompt do its one job.