AI-generated property and product images: where editing becomes misleading advertising

AI tools can now remove a bin from a driveway, furnish an empty room, turn a mug photo into a styled scene or lift a dull sky. Those edits are fast and cheap. They also create a new kind of complaint: the buyer who arrives at the house or opens the parcel and says the picture did not match.

This guide is about the cross-cutting test that regulators use, and a simple way to apply it before you publish. I keep the how-to steps for specific edits in other pages and link to them. Laws and platform rules differ by country and change often, so treat this as a map and check the current rule for your place and channel.

The test regulators use: net impression and materiality

The FTC, the US consumer protection agency, publishes an advertising guide for small business. It says ads are judged by the net impression of the words, phrases and pictures together, from the point of view of a reasonable consumer. It also says an ad can deceive through what it leaves out, and that you need a reasonable basis for a claim before the ad runs. Implied claims count as much as stated ones. A photo that implies a result is a claim.

In the UK, the Advertising Standards Authority (ASA) enforces the CAP Code, the rules for non-broadcast advertising. Three rules matter here: 3.1 says marketing must not materially mislead or be likely to; 3.3 says it must not omit material information; and 3.7 asks for documentary evidence to back objective claims before you publish. In plain words: do not mislead on something that matters, do not leave out something that matters, and keep proof.

Both tests reduce to two questions you can ask about every AI edit:

1. Is it material? Would it change a buying decision? Room size, light, view, condition and fixtures matter. A slightly nicer shade of sky usually matters less, but a sky that hides a power line may not.
2. Would a buyer who sees the real thing feel misled? If the answer is yes, do not publish it, or label it so clearly that the buyer knows exactly what they are looking at.

The FTC has no statement written specifically for AI-altered or doctored photos that is confirmed here, and the text of its older deception policy statement is not quoted. So this page applies the general test, not a rule about AI images. If you need certainty for a campaign, ask a lawyer who knows advertising law in your country.

Property images: what buyers rely on

A listing photo does a job a viewing cannot: it decides whether anyone books a viewing. So the facts a buyer or renter takes from the photo are material. The ones I would never alter:

• Size and layout. Wide-angle distortion, removed walls, merged rooms, an enlarged garden.
• Furniture scale. Shrinking furniture to make a room look bigger, or using an outdated or selectively cropped image, creates the same false impression (a New Zealand law firm lists these as risks under its Fair Trading Act).
• Light and view. A new window, a replaced view, a sea that is not visible from the unit.
• Condition. Stains, cracks, damp, worn flooring, a dated kitchen. Retouching these out hides a defect.
• Fixtures and what is included. A pool, a fireplace, built-in wardrobes, appliances that are not staying.
• The surroundings. Roads, power lines, nearby buildings, construction.

Virtual staging, which means adding digital furniture to an empty room, is the best-known case and has the most detailed rules. In the United States, several local MLS (multiple listing service) bodies and a California law on digitally altered images in listings have disclosure rules. I do not repeat them here. See virtual staging a property: what you must disclose for the rules and sources, and AI virtual staging and listing media in Dubai for the local position there. Walk-through video has its own traps with warping and invented rooms, covered in turning property photos into video.

Product images: what buyers rely on

For a product, the buyer relies on the picture for colour, size, finish, material, quantity and what is in the box. Product photos are also where AI edits creep in without anyone noticing, because the tool "improves" the item.

• Colour and finish. A tool that makes leather look glossier or a fabric look richer changes what arrives.
• Size and scale. Props in a styled scene tell the buyer how big the item is. A mug next to a tiny plant looks bigger than next to a large one.
• Quantity and accessories. Show only what is in the box. A generated bowl of fruit next to a blender is fine as scenery, but it must not look like it is included.
• Features. An invented button, port, label, certification mark or logo is a claim. Generated text on packaging is a common source of errors.
• People. A generated person using the product implies a real user or a real result. More on that in the platform table below.

An AI scene background is usually fine if the product stays true to the real item and the scene does not suggest size, performance or inclusions that are untrue. For removing one object, the fair-versus-misleading line is covered in how to remove an object from a photo, and label warping when you animate a product is in turning a product photo into a video.

The edit ladder: restore, tidy, invent

Here is the tool I use. Every AI edit sits on one of three rungs. Ask the question at the right of the rung.

Rung	Examples	The question	Usual verdict
1. Restore	Fix colour so it matches the real item; correct exposure or white balance; straighten verticals; remove dust on the lens or sensor; reduce noise; upscale resolution	Does the edit move the picture closer to what the buyer will see in person?	Low risk. Keep the original.
2. Tidy	Remove a bin, a stray cable, a car, a price sticker; clean a background; replace a plain studio backdrop with a neutral one	Would a buyer feel misled when they arrive? Is the removed thing part of what they are buying or living with?	Medium. Fine for temporary clutter, risky for anything permanent (a pylon, a skip outside, a stain on the wall).
3. Invent	Add furniture, add a view, add a feature, repair damage, change size or material, add a generated person or a lifestyle scene that implies something	Does it show something that is not there or hide something that is?	High. Disclose clearly next to the image, or do not do it.

Two practical notes. First, rungs can blur: removing a bin is tidy, removing a stain on a wall that stays is a defect hidden. Second, the rung is decided by the effect on the buyer, not the tool. The same button can do all three.

Disclosure wording that sits next to the image

A disclosure that nobody sees does not help. Put it next to the image, on the same screen and in readable text, not only in a footer. Short, specific wording works better than a generic line. Templates to adapt:

• Staged room: "Furniture and decor in this image are digitally added. The room is empty in person. See the original photo."
• Cleaned photo: "Temporary items were digitally removed from this photo. Lighting and colour adjusted to match the room."
• Product scene: "Background is a generated scene for illustration. Only the item shown in the foreground is included."
• Generated person: "Person shown is computer-generated and not a customer."
• Sky or exterior change: "Sky digitally replaced. Real conditions will vary."

Offer the original. A line like "Original photo available on request" or a second image in the gallery helps trust and works as evidence. Whether a given label is enough depends on the law and platform, so check the current rule for the channel.

Platform and law rules in one table

This table collects the sources. Each row is a snapshot, so re-check the linked page before relying on it. For ad platform rules and a pre-flight checklist, read AI-generated ads and platform ad policy traps.

Place	What it says
Etsy, an online marketplace for handmade and vintage goods	Its listing image requirements (last updated 22 October 2026) say most listings need one unedited real photo. Other listing images may use AI if they are accurate.
Amazon, an online store	A Seller Central forum notice from late July 2026 says photorealistic AI-generated people in images or videos need the keyword contains-synthetic-performer in the dc:subject XMP metadata field. The wider scope of the rule is unclear to me, so check Seller Central.
TikTok, a short-video platform	Its support page says a label is required on all realistic AI-generated content, and that content made with TikTok's own AI effects or carrying C2PA content credentials is labelled automatically.
New York State	A law on synthetic performers in ads (signed 11 December 2025, effective 9 June 2026) requires disclosure when the advertiser knows a synthetic performer appears in the ad. Penalties are $1,000 for the first violation and $5,000 for each later one, per the bill text (New York Senate bill S8420-A), which covers a digital performer that gives the impression of a human who is not recognizable as any real person, and exempts audio-only ads and promotion of films, shows and games. The state's Department of State also issued an alert in November 2025 warning homebuyers about AI-generated home listings.
European Union	Article 50 of the EU AI Act applies from 2 August 2026. People who deploy deepfake-style image, audio or video must disclose that it is artificially generated or manipulated, at the latest at first exposure. This is a transparency rule and does not replace advertising law.
California, USA (property listings)	The San Diego MLS (a local listing service) says Assembly Bill 723 took effect on 1 January 2026. Altered images must be labelled and paired with the unaltered original. Basic edits such as brightness, colour correction and cropping do not count as altered, although it suggests disclosing when unsure. Its rules cover MLS listings, and agents must still comply in advertising outside the MLS. One photography site says October instead, so check the date with your MLS.
Wisconsin, USA (property ads)	Straight Arrow News reports a law requiring disclosure of technology-altered ads, including AI-made ones, that create a false or misleading impression, effective 1 January 2027. Check the statute itself before relying on this.
United Kingdom	The CAP Code rules above (3.1, 3.3 and 3.7) apply to non-broadcast ads, including product and property marketing.

For the wider question of when to tell clients and audiences about AI use, see should agencies disclose AI use.

Keep originals and an edit log

If a buyer complains, a regulator asks, or a platform challenges a listing, you want to answer quickly. CAP 3.7 and the FTC idea of a reasonable basis both point to the same habit: keep evidence. A simple file per image:

• The untouched original, with the date it was shot.
• The edited version.
• A one-line edit log: tool, what was changed, rung 1, 2 or 3, who approved.
• A side-by-side check, original beside edited, done before publishing. Two minutes. Look for anything that now looks bigger, cleaner, newer or more included than the real item.
• The disclosure text used, and where it was placed.

Store these with the job. The same discipline works for a studio managing hundreds of images, and an asset library makes it routine.

Local rules and where to go next

Countries and local bodies differ. For US property listings, California and several MLS organisations have specific rules; the staging page above lists them. For Dubai, the local requirements are in the Dubai page above. For the UK, start with the ASA's CAP Code section 3. For EU use, start with Article 50 and your national advertising body. If your work crosses borders, follow the strictest rule that applies to the places where the ad is shown.

Who should not AI-edit at all

• Anyone selling something whose condition is the point. Used goods, vintage items, damaged stock, rental properties where wear matters. Use real photos.
• Anyone with a regulated product. Food, cosmetics, supplements, medical devices and financial products have rules about what images can imply. Check with the regulator or a lawyer first.
• A listing with a known defect. If the roof leaks, do not paint the ceiling in post.
• Anyone who cannot keep originals. If you cannot show what the real thing looked like, do not edit.

A real photo beats an AI one for any claim about condition, size or colour. Competing tools and a photographer with a good lens are the better pick whenever accuracy is the product. The same limits apply to video, covered in when not to use AI video.

How I'd approach this

I would set one rule for the studio: rung 1 edits are fine with the original saved, rung 2 edits need a second look, rung 3 edits need a disclosure next to the image and the client's written agreement. Then add a two-minute side-by-side check to the publishing checklist and keep the edit log with the file. When a client pushes for "just make it look better", I would show them the ladder and let them choose the rung on purpose.

If you publish many listings or products and want the originals, edit log and approvals to live in one place, Blur Studio builds approval portals and shared asset libraries around the tools you already use. This page is information, not legal advice. For anything high stakes, ask an advertising lawyer in your country.