Asset Studio Product Images: Checking the Product Before the Scene

What Google's tools do with your product

Google Ads offers product-image generation from supplied photos and an AI image editor for changes to an existing scene. Both need the same scrutiny. Do not assume that a request to change the background leaves every buckle, label or seam untouched.

Feed imagery has an additional trap. The Google Ads image editor works on an Asset library copy of the Merchant Center image: edits do not update the feed, and subsequent feed changes do not refresh that copy. New packaging on your product page can therefore sit beside old packaging in an ad. Recheck the actual creative when the product changes.

If you want motion rather than a new scene, Google's Image-to-Animation tool is a separate option. Preview its output with the same product references; motion does not make a mismatch less important.

The check, product by product

Generated lighting, angles and surroundings are things the model made up. Check the details it had to reconstruct, and do not assume a simple product is safe.

A hypothetical: a skincare brand generates a bathroom-shelf scene for its 30 ml serum from one front-on photo. The scene is excellent and the bottle has a dropper cap; the SKU ships with a pump. Adding a photo that shows the pump and regenerating may fix it, and may not. If the second attempt is still wrong, the honest options are the real product photo placed on a background in your own design tool, or the real photo on its own. The product page, not the prompt, is the standard.

• The variant. Generate from the photo of the exact item the ad will sell, and compare the result with that feed row's image and title, not the range hero.
• Closures and hardware. Caps, pumps, zips, buckles and ports are where a model that saw one angle guesses.
• Colour and finish. Generated lighting shifts colour; compare against the product page image in the same tab.
• Pack contents and quantity. A generated scene can add a second unit or an accessory that is not in the box.
• Scale. Props and hands set the size of the product in a viewer's mind.
• Printed text and logos. Generated text is unreliable letter by letter; check every character or keep the label out of frame.
• Simple shapes too. A mug's handle, a candle's wick or a pot's rim can change. Familiarity makes these mistakes easier to overlook, not less misleading.

When to stop generating the product

For products whose value is in detail, jewellery, electronics with ports, patterned textiles, anything with printed text, treat generation as a way to draft a setting, not to produce the product. If a generated image cannot be made to match after a reasonable attempt, use the real photo. A composite you build yourself, with the actual product photo over a generated or stock background, is a non-generative path for the part that has to be exact.

Whichever path produced the image, the AI label question follows it. Google Ads Help says AI regulations in the European Union, India and New York require that ads with certain AI-generated or edited assets include disclosures, and offers an asset-level AI label setting that adds a disclosure in My Ad Center and visible overlays for campaigns targeting those regions. Google also says using the setting doesn't guarantee your compliance with specific regulations and to seek legal guidance. Which assets count as certain assets is a question for that guidance, not for this page.