AI product photos: background, retouching and the limits of truth

In an online store, the photo is the most important selling point. The customer can’t pick the product up, so they judge it by the picture. AI can turn a phone photo into a catalog-quality photo in a few minutes, but the same tool will easily draw in things the product doesn’t have.

In this guide I show where improving ends and misleading begins. You also get a simple workflow you can put in place today.

A simple rule: show what the customer will get

There is only one test. When the customer opens the package, the product should look the way it does in the photo. Anything that doesn’t break this promise is usually fine. Anything that changes it is a risk.

Before every edit, ask yourself: would a customer, seeing the photo and the product side by side, feel cheated? If so, don’t publish it.

What is fine

These edits don’t change the product, only its surroundings or the quality of the photo:

• Changing or removing the background. A white, light or neutral color instead of the mess in the workshop.
• Correcting light and colors so that they match the real product.
• Straightening, cropping and sharpening the shot.
• Removing dust, fingerprints and marks left by protective film that are not a feature of the product.
• A setting that shows scale, for example a mug on a table, if the mug itself is shown faithfully.
• A shadow under the product, so that it doesn’t “float” in the air.

The key word is color. Many returns in online stores happen because the color on screen doesn’t match the real one. Compare the edited photo with the product in your hand, ideally in daylight.

What is misleading

This is where the line is. The following changes alter the impression of the product:

• Removing defects the product actually has: a scratch, a chip, discoloration, a dent. If it is a second-grade or outlet item, the defect must be visible or described.
• Changing the color, texture or finish, for example “boosting” the gloss or the depth of the wood grain.
• Adding elements that aren’t included: extra accessories, plants, furniture, a higher-end model.
• Changing proportions and dimensions, including accidentally when stretching the frame.
• A “before and after” effect the product doesn’t deliver: a cosmetic, a cleaning product or a paint shown with a perfect result you won’t achieve under normal conditions.
• A photo generated from scratch presented as a photo of the real product.

The general ban on unfair market practices comes from the Act on Counteracting Unfair Market Practices (ustawa o przeciwdziałaniu nieuczciwym praktykom rynkowym). Under Art. 5 of this Act, a misleading action may include, among other things, spreading false information, or true information in a way that may mislead, if it may cause the customer to make a decision they would not otherwise have made. A photo is such information. Leave the precise assessment of a specific case to a lawyer, but common sense is enough for most cases.

Table: a quick cheat sheet

Step 1: take a good starting photo

AI improves photos, but it won’t replace good source material. A phone is perfectly enough if you take care of a few things:

1. Place the product by a window, away from direct sunlight.
2. Use a light sheet of paper or a wall as the background.
3. Take 3–5 shots: front, side, a detail, scale (e.g., next to a hand or a ruler).
4. Clean the lens and the product.

Keep the starting photo without any filters. That’s your original.

Step 2: edit carefully and in small steps

It’s safest to give AI one task at a time. Instead of “make a nice photo”, write something specific:

Remove the background and replace it with a solid light gray. Don’t change the shape, color, lettering or texture of the product. Add a soft shadow underneath.

After each edit, compare the result with the original at high zoom. AI tools tend to distort small details: the lettering on the label, the number of buttons, the shape of a handle, a fabric pattern. Pay particular attention to text and logos, because that’s what most often gets mangled.

Step 3: check compliance with platform rules

Platforms have their own image requirements. In its Merchant Center help, Google requires the image to show the product being sold, without promotional overlays or watermarks. Sales platforms have similar rules, for example Allegro in its rules on thumbnails, meaning the main photos of listings. Check the current rules of the platform you sell on before uploading edited photos. If you run a feed for Google Merchant, incorrect photos can lead to products being rejected.

Step 4: label AI photos where necessary

Not every edit needs a caption. Smoothing the background or correcting the light is ordinary editing, like in a graphics program. It’s different when an image is entirely generated or realistically altered so that it could pass for a real photo. That is where Art. 50 of the AI Act comes in, which covers labeling such content. I describe how to do this in practice in the guide on labeling AI content on a business website.

A simple caption under the photo is enough:

Visualization generated with the help of AI. The actual product may differ slightly.

I apply this rule: when in doubt, label it. A caption takes a few seconds and builds trust.

Step 5: an archive of originals

Create a separate folder on a drive or in the cloud and keep these in it:

• the original photo from the phone,
• the edited version,
• the date and a short note on what was changed (e.g., “background, light”).

Name files consistently, for example blue-mug_original.jpg and blue-mug_web.jpg. The original will come in handy when a customer files a complaint because the product doesn’t match its description, when you need to go back to the unedited version, or when you switch tools and want to redo the photo. It’s worth including this archive in your regular backups.

Project photos: extra caution

Service businesses, construction companies for example, show photos of completed projects. AI can help here with framing and light, but don’t draw in elements that weren’t built. If you show a visualization before the work starts, label it clearly. I write more about the limits of visualizations in the guide on AI bathroom visualizations. Also remember to get the client’s consent to publish photos of their home.

Summary

AI works great for background, light and framing, but it must not change what the customer will get. Take a photo with your phone, edit in small steps, compare with the original, label generated images and keep an archive.

If you want to put the photos in your store in order or check whether your product presentation complies with the law, write to me. I can help with graphics and product photos, and in your store I’ll check legal compliance. I reply within 24 hours.