The window peeper prompt
The window peeper prompt has to do five things at once: keep the stare on the lens, keep the camera outside the glass, keep the food free of brands, time the curtain, and keep the person small and recognizable in the frame for the first 3 seconds. That is a lot to ask of one paragraph, so on Window Peeper AI it is already written and tested. Make the video from one photo.
What the prompt has to solve
People searching for an AI window peeper prompt usually expect a block of text. The text is the smallest part of the problem. These are the five things any version has to get right.
- The stare stays on the lens. The joke is about 6 seconds of direct eye contact, from about 5 to 11 seconds. Video models like to look away, blink or glance at the food. A prompt has to hold the gaze without making it look frozen.
- The camera stays outside the glass. The shot is made from outside the building, looking in. Models tend to drift the camera through the window into the room, which turns a creepy comedy into an ordinary dinner scene.
- The food has no brand. Burgers and bags come out with invented logos. That is a risk for you as a poster and a problem for any platform review, so the food must be plain.
- The curtain lands on time. The stand-up starts at about 11 seconds and the curtain closes between 12 and 14. If it comes early the stare is cut short, and if it comes late the clip runs out before the punchline.
- The person is small, yet still you. For the first 3 seconds the person is a small figure behind glass. The face has few pixels, so identity must be pinned without making the figure larger than the shot allows.
The beats come from a real shot sheet: building, window, bite, stare, stand and curtain. See the six beats of the trend for the timings.
Why copied prompts fail
The typical failures people post about each have a cause that is not in the words.
- Wrong face. A pasted prompt cannot carry your photo. Without a tuned way of using the photo, the model invents a person who resembles nobody.
- Wrong timing. A sentence like "stares for a few seconds" is read loosely. The model decides how long, and the curtain moves with it. Timing needs a reference, not an adjective.
- The camera goes inside. A single mention of "window" is not enough to keep the lens outside. It has to be a rule repeated in the right place.
- Different model, different result. A prompt tuned on one model version behaves differently on another. Copying text from a tutorial for a different tier gives you someone else's luck.
- Paid retries. Every failed attempt costs credits at a video model, so trial and error adds up quickly.
Anyone who tells you the right words alone make the clip is leaving out the other inputs. Window Peeper AI does not publish its prompt, and it does not need you to paste one.
What is built in here
Each of the 6 scenes has its own fixed direction, motion reference and settings. You supply only the photo (two on Couple Dinner), pick a scene and a format, and the shot list does the rest. A Standard 720p render costs 10 credits and takes a few minutes. If a render drifts, run it again; failed or blocked renders return their credits automatically.
Limits are real. The person is small in the first 3 seconds, so a sharp front-facing photo matters, and there is no free tier. Read how to make the video in four steps, or compare with a CapCut template.