
How an ad is made from one photo walks the whole flow; this page is about the photos themselves.
Home page, Ad Video tab. Add images opens the file picker; Add another image adds more, up to four; Remove image takes one out. The note under the box says the rules: up to four images — product, shopfront, dish, car, home or app screenshot — JPG or PNG — make sure you own the rights.

The thing being advertised fills most of the frame and is in focus. A shopfront sign, the dish itself, the car, the room, the product on a table: anything the reading can recognise. If it cannot, the message says so — We couldn’t work out what is being advertised here — try clearer images — and offers Choose another. Nothing is charged for a refused image.
The first image is the featured product: the one the ad is built around and the one the end card shows. The others are treated as angles, details or the place it is used, and become the sets the shots are made in. Put the hero shot first.
A restaurant photo has diners; a car photo has a driver. That is fine: the reading notes it, and the composer says People will be removed. The place or the product is kept, the people are taken out, and the ad’s own cast is generated instead — so nobody who happened to be in your photo ends up in your ad.
Screenshots are the exception. A screenshot with a real face is refused (This screenshot has a real face in it. Please use one that shows only the interface), because editing the face out would smear the interface with it. Use a clean screen.
An app or website screenshot is used as it is; it is never redrawn, so the interface stays readable. It must be roughly a phone or computer screen shape: a full-page capture is far taller than any video frame and is refused with This screenshot is too tall or too wide — crop it. Crop to one screen and upload again. Ads for apps and websites.
Press Make my ad and the composer says Reading your image…. A moment later a dialog says AI has read your product and shows what it understood: the category (Product, Food & drink, Apparel, Vehicle, Store / property, App, Website, Service) and a Product or place name field you can correct. If the reading has misunderstood the thing entirely, Pick different images. The one line you typed under the images — what it is and who it is for — is read too, and wins over the picture when they disagree. Nothing is charged until you choose how to make the ad. The two ways to make it.
A name printed on a shopfront, a package or a car is read as the brand and printed on the end card. Video models do not write legible text, so the sign in the picture is not something to rely on; the end card is where the name is delivered. The end card.
| Photo | Works? | Note |
|---|---|---|
| Product on a plain background | Yes | Best as the first image |
| Shopfront or restaurant with customers | Yes | People are removed; the place is kept |
| Dish, drink, garment, car, room | Yes | Fill the frame, in focus |
| App or website screenshot | Yes, if it is one screen | Never redrawn; no faces |
| Full-page screenshot | No | Crop to a screen shape first |
| Screenshot with a real face | No | Use a clean screen |
| Blurry, tiny, or ambiguous | No | Try clearer images |
Up to four, JPG or PNG, each under 10 MB.
Fine. People are removed and the place is kept.
Either it has a real face in it, or it is a full-page capture. Use a clean, one-screen crop.
Yes. The dialog after the reading has a name field; correct it before choosing how to make the ad.
No. Nothing is charged until you choose one-click or expert mode.
One clear photo is enough; three more show the angles and the place.
Updated September 26, 2026
Pricing·Novel to Video·Guides·Samples·Support·Privacy·Terms·Legal·Contact·© 2026 SceneMixer