# How to Make a Caught at the Window AI Ad

> Make a Caught at the Window AI ad with the right product and person photos, a believable eating action, and checks before you post the fictional scene.

- URL: https://trendai.video/blog/caught-at-the-window-ai-ad-guide
- Published: 2026-10-17
- Category: UGC & AI Ads
- Tags: AI commercial, caught at the window, food ad
- Author: Trendai.video

To make a Caught at the Window AI ad, start with a food or product that reads from across a room, then decide who gets caught enjoying it. The Trendai template moves toward a lit apartment window, lands on a mid-bite reaction and ends when the person shuts the curtains. You got caught.

## The window is the hook, not the packaging

The [Caught at the Window template](https://trendai.video/trends/caught-at-the-window-ad) stages a camera approach from outside the apartment. Inside, a person eats or drinks, notices the lens, gives it an annoyed look, sets the food down and closes the curtains. That movement gives a small food brand a story without a voiceover explaining the product. You're interrupting somebody's private snack. That is the bit.

The live page describes a 12-second vertical video at 9:16 and 720p. It breaks the reference scene into an approach, an eating beat, a reaction, a curtain close and a short hold on the shut curtains. Those beats matter more than a perfect wrapper. If the camera hasn't reached the window before your viewer scrolls, the joke never gets a chance.

This is an independently made fictional ad scene, not documentary footage of someone being watched in their apartment. Don't caption it as if you secretly filmed a customer. For a shop, the cleaner angle is a playful confession: “When the last serving was supposed to be for tomorrow.” It gives the annoyed look a reason and keeps the product in the story rather than pretending this is a real surveillance clip.

For a product-first clip without a window reveal, the [fashion accessory UGC script guide](https://trendai.video/blog/fashion-accessory-ugc-ad-script) gives the item a different opening shot.

## Cast the two optional uploads correctly

The [live creation form](https://trendai.video/trends/caught-at-the-window-ad) asks for an optional food or product photo first, then an optional photo of one adult who agreed to appear. Both are optional. Without them, the currently selected defaults are Nadia as the character and pasta as the food; the form also offers other built-in character and food choices. The upload fields don't ask for two people, and the first slot is not a background image.

For the first slot, shoot the actual item you want noticed. A plate of pasta can survive the camera distance because its shape is obvious; a tiny label on a bottle probably can't. If you're using packaged food, put the whole package against a plain background with enough space around its edges. Skip a busy shelf photo where the generator has to guess which product is yours. Tiny text is risky. The form warns that small print can change, so never make ingredient claims or a discount code depend on AI-rendered package text.

The second upload is the star at the window. Pick a clear, well-lit photo of one consenting adult, with the face unobstructed and enough of the upper body visible for a believable eating pose. A sunglasses selfie or a photo with a second face makes the reaction harder to read. Ask the person before using their likeness in an ad, even if they're a friend who already posed for your store account. Consent to a photo isn't automatically consent to a fictional commercial.

No product photo? The food selector can carry the prop. No person photo? Use a built-in adult character instead of borrowing somebody's face. If your brand sells a nonfood item, be honest about the scene's limits: a “mid-bite” setup may be a bad fit for a phone case, regardless of how sharp the product photo is.

## Set an action the prop can actually perform

Under Customize, the form offers an eating or drinking action and a music setting. It currently shows “Match the food” for the action and music on; its food options include pasta, burger, soup or iced coffee. Pasta being twirled on a fork reads differently from sipping a drink. If you describe your own item, look at the chosen action before generating. A fork near a sealed cup would make the interruption funny for the wrong reason.

Make one simple decision about what the person is doing before they're spotted. For a burger, let the bite be the point. For soup, the spoon should be recognizable even on a phone screen. Need a product demo? This window gag may bury the feature you're trying to sell. A plain product shot elsewhere in your edit can show accurate details after the joke, but don't imply the template itself adds a logo or closing text: its page says those aren't available yet.

The form says music plays under the scene and dips when the character notices the camera. That's a configuration and source-scene description, not a promise that every personalized export will sound identical. The generation note says the video is made first and the soundtrack is added afterward. Check the downloaded result for the music transition and curtain sounds, plus any awkward silence before posting; if you choose music off, review the generated scene sound instead. Add only audio you have the rights to use in the editor where you'll publish.

## Review the reveal before you write the caption

Watch the export once with the sound off. Can you tell what the person is eating before they spot the camera? Does their expression change at the right moment, and do the curtains actually cover the window by the end? The template's poster shows a woman holding a fork above a plate of pasta at a nighttime window. It's a source still, not evidence that your uploaded product or face will render accurately.

Pause. Look at the item and the face. Compare the product shape against your reference, then check whether the person still looks like the consenting adult you uploaded. Packaging and likeness are generation goals, not guarantees. If a name or label turns into nonsense, don't leave it in a paid ad hoping nobody notices. Rework the input photo or avoid relying on readable text within the shot.

Keep the apartment window in frame if you resize the export for another placement. A tight crop of only the face loses the outside-looking-in setup; a crop of only the food loses the caught reaction. It is already vertical 9:16. You shouldn't need to turn a widescreen reference into a vertical post. If you add your own caption, leave the window and curtain line visible underneath it.

For the actual post, try a caption that sets up the interruption rather than naming every feature. “POV: you said you were saving the leftovers” lets the curtain close be the punch line. A restaurant could pair that with a real menu link or accurate offer in its own post copy, outside the generated frame. Check the final caption against the actual item shown, disclose the AI-made portrayal where the platform requires it and never present the scene as a real customer's testimonial. If you need a different scene rather than a window gag, browse the [trends gallery](https://trendai.video/trends) before forcing this one to sell something it can't show.