EU AI Act & Product Photography: AI Image Labeling Rules for E-commerce Brands (2026)

EU AI Act Product Photography: Complete Labeling Rules

The question I get most from clients since August is short: do my images need an AI label now? The honest answer comes down to one question. Did AI clean up a real photo, or did it change the product? After years of QC on ecommerce images, I can tell you that this one question sorts almost every file on my desk. Below, I walk you through the EU AI Act product photography rules the way I explain them to clients, using the European Commission's own guidelines. I'll also show you where our photo editing services fit, and where they don't. 

One quick note first: I'm a photo editor, not a lawyer, so treat this as practical guidance, not legal advice.

What the EU AI Act Product Photography Rules Mean for Your Catalog

The rule you're dealing with is Article 50 of the AI Act. It started applying on August 2, 2026, and the Commission explains it in its Article 50 guidelines. The rule splits the work between two players. The company that makes the AI tool has one job, and you, the brand using that tool for business, have another.

That December date isn't for you. It only gives older AI tools extra time to add their hidden marks, and your side of the rule is already live.

The fines are worth taking seriously: up to €15 million or 3% of worldwide turnover. Small businesses face the lower of the two. So if your store turns over €2 million a year, your ceiling is 3% × €2,000,000 = €60,000. That's a lot of retouching budget.

Who Has to Add the Label: You, Your AI Tool or Your Agency?

If you sell for a living and use AI images, the label is on you. The guidelines treat any activity that brings you money on a regular basis as professional use, so a one-person online store counts.

Hiring someone doesn't pass the job along, either. If a freelancer runs the AI tool for you under your brief, you're still responsible. The only way out is to genuinely hand over the decision. A brand that commissions an agency and has no say in whether or how the agency uses AI isn't treated as the one using it.

This is how I explain it to clients. If your brief says "put it on an AI beach," you made the call. Write the labeling step into the brief or the contract, and don't assume someone else will handle it.

Retouching vs AI Generation: Where I Draw the Line on the Desk

This part matters most for anyone who edits product photos. The rules exempt "standard editing," which the Commission describes as preparing existing content for publication without creating anything new. Its examples line up almost exactly with how I sort a job.

Here's how the line looks in practice. When I knock a photographed handbag out onto pure white, nothing new enters the frame, so that's standard editing. If a brief asks us to add a gold chain strap the bag doesn't have, the image now shows a detail the camera never captured.

My QC habit makes this easy to check. I compare every edited file with the raw capture, not with the brief. Then I ask one question: did anything about the product change that the camera didn't record? The tool doesn't decide the answer; the change does. That applies to our own work too, since we use AI-assisted tools alongside manual Photoshop editing.

When EU AI Act Ecommerce Images Count as Deep Fakes

Most EU AI Act ecommerce images I see never get close to this line. Still, "deep fake" makes people think of fake politicians, and the EU's definition is much wider. It covers products. An image counts when it passes four tests:

1. It resembles a subject.

2. The subject exists or plausibly could.

3. The subject is a person, object, place, entity or event.

4. The image would falsely look authentic or truthful.

The guidelines say "objects" includes consumer goods. They even list an AI product image in an ad or on packaging that could mislead about how the product really looks, what it's like or how it's used. Making the product look better than it is counts.

There's good news on the other side of that list. A real product shown on an AI-generated background is fine, as long as the ad doesn't mislead about the product itself. Swapping a background for clearly aesthetic reasons is treated as minor. Photorealism alone doesn't decide anything, because context does.

What About AI Models and Virtual Influencers?

This is where I see brands take the biggest risk. The guidelines name a video of a realistic synthetic influencer testing a real product as a deep fake, and they say it doesn't get the lighter treatment reserved for creative work. My reading is that a still photo of a photorealistic AI model wearing your jacket sits on the same side. Amazon is already strict here: its seller announcement requires a metadata tag on images with photorealistic AI-generated people. Label these images clearly, every time.

How to Label EU AI Act Ecommerce Images Without Breaking Marketplace Rules

The EU label has one job: your shopper has to see it. A hidden watermark won't do, because nobody can read it without special tools. A line in your terms of use won't do either. The label needs to be clear, sit on the image, and appear the first time someone sees it. The Commission's Code of Practice even offers official EU icons you can use.

Then the marketplaces add their own AI generated image label requirements, and those live in the file's metadata.

Google's Merchant Center policy lists the tag values it accepts.

The production problem I run into most is metadata getting lost. Someone writes the tag into the master file, exports a web copy, and the tag disappears, because many export presets strip metadata. So before any tagged file leaves our desk, I open the final export in a metadata viewer and check that the tag survived. It's a quick check, and it can save you a rejected listing.

AI Transparency Product Photos: When I Tell Clients to Shoot the Real Thing

When brands ask me about AI transparency product photos, I don't tell them to avoid AI. A labeled AI lifestyle scene works fine for a campaign banner or a social post that sells a mood.

I draw a hard line on images that have to prove the product. Your main listing image, fit shots, texture close-ups and anything showing how the product works should be real photos with standard retouching. Shoppers get what they saw, and you skip the label question entirely.

That's the work our team does every day. Rates start at $0.30/photo, with bulk discounts at higher volumes. Our photo editing pricing breaks down the exact rate for each service.

AI Image Labeling 2026: My Pre-Publish Checklist

I run this AI image labeling checklist on every batch headed to EU shoppers, right alongside the marketplace checks we handle in our Amazon product photo editing work:

• Compare each edited file with the raw capture, and note every change the camera didn't record.

• Treat background removal, color correction and cropping as standard editing.

• Flag any added, removed or replaced product detail for a second look.

• Label any image that could mislead a shopper about how the product looks, what it's like or how it's used.

• Put the label on the image itself, where shoppers see it first.

• Check that the Amazon or Google metadata tag survived export before you upload.

Final Advice

If I could give you one habit, it would be this one. At the brief stage, decide which images have to prove the product and which ones only set a mood. Shoot and retouch the first group. Generate the second group if you want, and plan its label from day one. That keeps you on the right side of the EU AI Act product photography rules without slowing your launches.

If you're not sure where an edit falls, send us the raw file with your brief. We recommend following our checklist on how to prepare photos for high-end retouching. With the RAW files, color references and editing rules it covers, we can tell you whether the change stays within standard editing.