Window Dance AI Video: the lit-window dance, the caught grin, the curtain, explained

A window dance AI video is a 12-second vertical clip shot as if from the apartment across the street at night: a dark brick block, one warm lit window, and the person in your photo dancing wildly inside next to a small pink speaker, completely unaware [reported]. Then they notice the camera, flash a guilty grin, and pull the curtain shut [reported]. The whole thing is made from a single photo — nothing is filmed.

It is the second wave of the Window Peeper meme. The first wave, around October 20, 2026, was someone across the street caught eating a burger at the window [reported]. On October 22 a TikTok account posted a Window Dance CapCut template with an original sound, and it passed a million views [reported]. A second template followed on October 24, and pet takes appeared on October 25 [reported]. The AI generator wave — dedicated tools and first-and-last-frame routes — started around October 30 [reported].

The trend travels under two names: “window dance” for the clip, and “Window Peeper Dance” for the CapCut template [reported]. The AI-video phrasing is newer: on November 6, 2026, Google autocomplete returned no suggestions for “window dance ai video” or “window dance ai” [reported], so this page covers the trend under its established names instead of betting on one spelling.

Every version that works hits the same four beats in about 12 seconds [reported]. Knowing the beats matters: if your generation breaks, it breaks at a specific beat, and that tells you what to fix.

0–2 seconds — the push-in. A dark brick apartment block at night, then the camera pushes in on the one lit window. The window is the whole establishing shot: if the building is bright or several windows are lit, the joke has nowhere to land.

2–6 seconds — the dance. Inside the window, your photo’s person dances energetically next to a small speaker, seen from the knees up, completely unaware of the camera [reported]. This is the longest beat and the one the template is judged on — the dance has to read as joyful, not posed.

6–8 seconds — the caught smile. They notice the camera and flash a guilty grin. The beat is short and it is the whole joke: the shift from unaware to caught is what people share. If the model skips it, the clip is just a person dancing in a window.

8–12 seconds — the curtain. They pull the curtain shut and the window goes dark [reported]. One published shot list times the pull at 8–10 seconds [reported]; the clip holds on the shut curtain to about 12. Do not let the model reopen the window or cut back to the building — the dark window is the ending.

The joke is being caught. Everyone has danced alone in a room; the clip just adds a witness across the street. That is why the caught smile, not the dance, is the beat people remember — and why the curtain has to close on it.

The template has one parameter — who is in the window — and that parameter is a variant machine. Any person fits the same lit-window scene, and so does any pet: the pet takes that appeared on October 25 [reported] need no new prompt, just a different photo. Same mechanism as the tiny-desk and glowing-veins templates — one fixed scene, one swappable subject.

The second accelerator is that the CapCut template only swaps a face. It keeps someone else’s window, room, clothes and timing [reported], so anyone who wants their own room, their own clothes, or a pet has to leave the template and use a generator. That gap is what the AI generator wave from October 30 is selling: the whole scene, filmed around your photo [reported].

The honest clock: the CapCut template dates to October 22, and this page ships on day fifteen. That is past the usual two-week window for a variant page — but the AI generator wave only started around October 30, seven days ago, and that wave is what this page is actually about. The durable part is not this window, it is the mechanism: one fixed voyeur shot, one swappable subject, four beats.

There are three routes, and the choice is the page. The CapCut route is free and fast; the generator routes cost money and give you the whole scene. Pick by what you need, not by what is newest.

Route 1 — the CapCut template (free). Search “Window Dance” in CapCut, or tap the template link on any #windowdance video [reported]. Drop in a photo of your face and export. It is free and fast [reported]. The catch: it only swaps a face onto the original clip. You keep someone else’s window, room, clothes and timing, and the swap can look pasted on when the head turns [reported]. It also only works with human faces, so no pet version [reported]. The template is also listed as “Window Peeper Dance” [reported].

Route 2 — the dedicated generator ($9.99). Upload one clear photo of one person or one pet, face visible; a free photo check runs before you pay [reported]. About 3–5 minutes later you get a 12-second 9:16 720p MP4, plus a GIF, with no watermark and an original AI-generated beat and room sound [reported]. One video is $9.99, three are $19.99, no subscription [reported]. There is no free trial, and every video costs real GPU time — but if a generation fails, the credit goes back automatically [reported]. No app, no sign-up; you pay as a guest and get the video by email [reported].

Route 3 — first and last frame (credits). Design the first frame (you in the lit window) and the last frame (the curtain pulled shut) from your photo, then animate between them [reported]. On one tool the animation step costs 280 credits for 10 seconds, with credit packs starting at $9.90 for 250 credits [reported]. The same tool’s own remixes were cut from 10-second Seedance 2.0 Fast originals [reported]. This route gives you the most control over the room — you design both frames — and the most ways for the model to drift between them.

If you use a generic image-to-video generator instead, the condensed prompt below is our wording of the circulating four-beat structure, not a copy of any single prompt site’s text. The beats, the timing, and the rules are the same; the phrasing is ours.

Create a 12-second ultra-realistic vertical 9:16 video, filmed as if secretly shot from the apartment across the street at night. Start on a dark brick apartment block with one warm lit window; the camera pushes in to that window.

Inside, the person in the uploaded photo dances energetically next to a small pink speaker, seen from the knees up, completely unaware of the camera. Midway, they notice the camera and flash a guilty grin. Then they pull the curtain shut, and the window goes dark.

Keep the person's face, hair and clothes from the photo throughout. One continuous shot. No cuts, no captions, no extra people, no second window lit.

Sound: the dedicated generator ships its own beat [reported]. To use the trending sound instead, find any #windowdance video on TikTok, tap the sound, and choose “Use this sound” when you upload your MP4 [reported].

This is an early bet, not a proven search term. The CapCut template is 15 days old, past the usual two-week window for a variant page; the AI generator wave is 7 days old. Google autocomplete returned no suggestions for “window dance ai video” or “window dance ai” on November 6, 2026 [reported]. The trend is real; the English demand for the AI-video phrasing is not yet established.

The head term already has a dedicated tool. windowdance.com holds the dedicated-tool position for the head term [reported], so this page is a guide, not a tool pitch — and it does not expect to outrank the tool on its own name. The room is in the long tail: pet version, CapCut template, free-online variants.

The CapCut swap looks pasted on. When the head turns, the face swap can look pasted on [reported], and you are stuck with someone else’s room, clothes and timing. If the room matters to you, that is the reason to pay for a generator instead.

One guide is written by a toolmaker. The Medium “3 ways tested” guide discloses up front that its author built Window Dance AI, one of the tools it compares [reported]. Its facts check out against the tool’s own pages, but read its comparison as the toolmaker’s own.

No verified fully free full-scene route. The CapCut template is free, but it is a face swap, not a full scene [reported]. Every full-scene generation costs money — $9.99 a video on the dedicated tool, credits on the frame route [reported]. Whether any generic generator offers free credits for this workflow is unverified [unknown].

The joke is being watched — keep it fictional. The clip works because it looks secretly filmed. Never film real people through their windows: the dedicated tool’s own line is that the video is AI-generated, nothing is filmed, and real people should not be filmed through windows [reported]. The voyeur angle is the joke; the consent line is real.