Operations
AI product images: what can change and who approves
AI product images: what generative AI can change in a product photo, what the law won't let it change and where your team approves. A ShopGuru guide.
AI product images save studio sessions, but they raise a question traditional photography rarely did: is what you see in the image still the product the customer will receive? Here we separate what generative AI can change, what the law won't let it change and where the store's team decides.
Article summary
- AI product images start from a real photo of the product and change what is around it: the background, the light and the setting.
- Generative AI helps most in catalogues with many items, where redoing backgrounds and settings by hand can take weeks.
- A generated image that shows the product in another colour, another shape or with accessories that aren't in the box can be a misleading commercial practice under the EU Unfair Commercial Practices Directive.
- Since 2 August 2026, Article 50 of the AI Act requires whoever publishes an AI-generated "deep fake" to disclose that it was generated; for a product image, that depends mainly on whether the image could mislead about the product.
- The ShopGuru agent prepares generated images from the store's own photos, with the product cut out of the original and left unchanged.
- Every generated image reaches the app as a proposal, next to the original photo, and is only published with the approval of someone on the store's team, recorded with a name and a time.
- A 30-minute call with a ShopGuru engineer is enough to find out which catalogue photos would gain from generated images.
What generative AI does well in a product photo
A product photo has two parts: the product, which is what the customer buys, and the setting (background, shadow, light), which is there to show it well. Generative AI pays off in the setting:
- Putting the whole catalogue on the same background, when the photos came from different suppliers.
- Creating a setting for the same photo: the boots on a wooden floor, the lamp in a living room.
- Preparing a seasonal version of the main image without a new photo shoot.
In stores with hundreds of items, the team stops editing images one by one and starts reviewing proposals.
"AI can change everything around the product. The product stays as it is." ShopGuru
The line not to cross: the product in the image is the product in the box
Our position: in a product image, AI works on the setting and leaves the product alone. Image generators don't know where one ends and the other begins. A prompt like "warm late-afternoon light" can lighten the leather, stretch the sole or add a lace the model doesn't have.
In the EU, unfair commercial practices are covered by Directive 2005/29/EC, transposed in Portugal by Decree-Law 57/2008. A practice is misleading when it gives false information, or information that deceives, about the main characteristics of the product.
An image showing another colour or another size is information about the product. And "the colour is different from the photo" is a return reason every store owner knows.
"An image that sells the wrong product sells a return." ShopGuru
The cases that need the most attention:
| What changes in the image | Why it's a risk |
|---|---|
| The product's colour or texture | In clothing, footwear and home decor, it's often what decides the purchase |
| Proportion or scale | A vase in a generated setting can look bigger than it is |
| Objects that aren't in the box | The customer may think the cushion comes with the sofa |
| Text, labels and logos | Generators distort lettering and can invent brands |
| People wearing the product | A generated body can change how a garment hangs, and the customer decides on fit |
What the AI Act asks of AI product images
Regulation (EU) 2024/1689, known as the AI Act, has two rules in Article 50 that touch generated images, applicable since 2 August 2026. The full timeline, after the Omnibus amendments, is in our article on what the AI Act and the GDPR ask of an AI agent.
The first is for whoever provides the image generator: paragraph 2 requires generated content to be marked in a machine-readable format. Regulation (EU) 2026/1744 gave systems placed on the market before 2 August 2026 until 2 December 2026 to comply.
The second is for whoever uses the system and publishes the result. Paragraph 4 requires disclosure that the image was generated or manipulated when it is a "deep fake": AI-generated or manipulated content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. The European Commission's guidelines on Article 50, published on 20 July 2026, include consumer goods among "objects".
The same guidelines give pointers for product images. A real product against a generated background tends not to be a deep fake as long as the image doesn't mislead about the product's appearance, characteristics or use. An image that makes the product look different from the real one, or better than it is, appears among the examples of deep fakes.
Our reading is that the AI Act's line largely matches the rule that the product in the image is the product in the box, but each image is assessed case by case. Our recommendation: when in doubt, the product page says the image was generated. How to flag it is the store's decision, and it's worth checking with a lawyer.
"When in doubt, the product page says the image was generated. The customer loses nothing by knowing." ShopGuru
How the agent prepares product images and where your team approves
The ShopGuru agent starts from the product photos the store already has. It cuts the product out of the original photo and generates a studio background or a setting around it. The product isn't redrawn: the background and the light around it change.
Every generated image reaches the app as a proposal, next to the original, for your team to compare. Your team approves, edits or declines, and the decision is recorded with a name and a time. Declining is a normal decision, and the agent only proposes the same thing again if the data changes.
An example of a review in a shoe store
Fictional example. In a shoe store, the agent prepared two generated images for the Serra Boots, from the photo the store already used: one with a studio background and one on a wooden floor. Sónia, on the store's team, compared both with the original.
- Studio background: the colour of the leather and the shape of the sole match the original. Approved.
- Wooden floor: the warm light of the setting makes the leather look two shades lighter. Declined, with the reason written in the log.
The first went onto the product page with a note that the image was generated, as the store decided to do for all of them.
In the review, your team looks at the same things it would in a photo shoot: colour, shape, scale, what appears next to the product and whether the lettering on the label can be read.
When generating product images isn't worth it
For some products, a generated image adds little or risks too much:
- One-off or second-hand items, where the customer buys that exact piece, with the marks it has.
- Products where the exact colour is the decision, such as paint, yarn or make-up; there, a calibrated photo is worth more than any setting.
- Blurry or badly lit original photos: AI improves the setting, but the product is still as badly photographed as it was.
The image doesn't answer everything either. Measurements, materials and care are settled in the text, and we explain how in the article on product pages that answer your customers' questions.
Frequently asked questions
What does the agent start from to prepare product images?
From the product photos the store already has. The agent cuts the product out of the original and generates the background or setting around it, so that the product in the image is the one in the real photo.
Do I have to tell customers that a product image was generated with AI?
Article 50(4) of the AI Act requires disclosure of deep fakes, generated or manipulated images that could pass for authentic. Whether a given product image is covered depends on the case; how to flag it is the store's decision, and it's worth checking with a lawyer.
Who is responsible if a generated image misleads the customer?
The image is published by the store, which is the one selling the product. That's why generated images arrive as proposals: someone on your team compares them with the original before publishing. The store owner chooses the autonomy level for each type of action.
Where to start
The first step is to look at the catalogue photos and separate those that would gain from a background or a setting from those that should stay as they are. That's what we do in a first 30-minute call with a ShopGuru engineer. Book a call.
About this article
Written by ShopGuru's engineers based on how the service is designed. Published on 21 October 2026. The store, the product and the names in the example are fictional. It reflects the legislation in force on that date: Regulation (EU) 2024/1689 (Articles 3 and 50), as amended by Regulation (EU) 2026/1744, the European Commission's guidelines on Article 50 of 20 July 2026, Portuguese Decree-Law 57/2008 and Directive 2005/29/EC. This article is for information only and is not legal advice.