AI product photography regulations 2026 for sellers

AI product photography regulations in 2026

TL;DR. Every AI product image rule in 2026 asks one of two questions: does the picture show the real product, and does the buyer know AI made it. August 2, 2026 is when the EU AI Act transparency obligations and the California AI Transparency Act both went live. That date has passed. One carve-out remains open, and the EU has published official icons for labelling AI content. Everything below is the country-by-country and marketplace-by-marketplace evidence behind those two sentences.

In September 2025, China’s AI labeling Measures went live and German market surveillance authorities opened the first investigations into AI-generated product imagery on major eCommerce platforms. The quiet phase of AI image regulation ended in one quarter. Background replacement, virtual model placement, lifestyle scene generation, all of it now runs from a browser tab. The rules around it now run from a courtroom.

No major market has banned AI-generated product images. Every jurisdiction now argues about disclosure and accuracy. AI product photography regulations differ from market to market, enough that selling globally means tracking five or six rule sets at once. This article maps where each major market stands in September 2026, what marketplaces enforce on top of that, and the small set of habits that keep a catalog compliant across all of them.

The two questions every rule asks

Strip away the legal language and every regulation collapses into two questions. Does the image show the product the buyer receives, and does the buyer know an AI tool made it. Accuracy is the older requirement, and consumer protection law already covers it everywhere. Transparency is the new one, and August 2026 turns on it.

Every section below is evidence for one of those two questions. The EU, California, China, and South Korea are mostly transparency stories. The UK and most marketplaces are mostly accuracy stories. The US federal posture is the messy hybrid in the middle. Read for the question you care about, not the country.

European Union

This is a transparency story. The EU AI Act eCommerce rules come down to one thing: disclosure. Existing consumer law already settled accuracy, and the new lever is mandatory disclosure. The EU AI Act is the most comprehensive AI regulation in force, and its transparency provisions for AI-generated content apply from August 2, 2026. That date has passed, and the Article 50 obligations now apply to every seller with EU-facing listings.

Article 50 sets up a dual obligation. AI tool providers must mark outputs in a machine-readable format, using mechanisms like C2PA metadata, imperceptible pixel-level watermarks, and provenance logging. Businesses that use those tools are deployers in the Act. They must tell consumers when a customer is looking at AI-generated or AI-manipulated content. The duty bites hardest on content that could pass for an ordinary photograph. On June 10, 2026, the Commission published the final Code of Practice on marking and labelling AI-generated content. The Code is voluntary. Signatories with a positive assessment can rely on it to show compliance with the Act’s marking and labelling rules, which makes it the closest thing to an official playbook. The Commission also published draft guidelines on Article 50 on May 8, 2026, with a targeted consultation that closed on June 3, 2026. Expect the interpretive detail to keep firming up through the year.

One obligation did slip, and the detail matters. The Digital Omnibus on AI entered into force on July 27, 2026. It delays the Article 50(2) marking obligation to December 2, 2026 for AI systems placed on the market before August 2, 2026. Everything else in Article 50 applies from August 2, 2026, including the deployer disclosure duty on sellers.

What triggers the deployer duty is worth pinning down, because it is narrower than most coverage suggests. Article 3(60) defines a deepfake as AI-generated or manipulated image, audio or video content resembling existing persons, objects, places or events that would falsely appear authentic. For a product catalog that draws a usable line. A folded shirt on a generated background does not trigger the Article 50(4) deployer disclosure. A generated human face that reads as a real, plausible person does. Apparel sellers using AI models are on the wrong side of that line by default. Our guide to AI photoshoots at catalog scale covers the likeness, release, and consent side of that problem.

Alongside the Code, the Commission released a free set of EU icons for labelling AI content. There are three. Each one ships as SVG and PNG, in black, white, and 50 percent transparency variants. All are free to download and need no attribution. This is what they look like and when each one applies.

The basic icon, the general-purpose mark for any content where AI played a part. It sits next to a short text label. Download SVG

The fully AI-generated icon, for content created entirely by AI with no human-made elements beyond the prompt. Download SVG

The partially AI-modified icon, for human-made content altered with AI. Most product imagery lands here, since a real product shot gets an AI background or scene. Download SVG

Using the icons is optional. The Article 50 labelling duty is not, and putting an icon on an image does not by itself make a listing compliant. The icons support the Article 50(4) disclosures, which cover deep fakes and AI-generated text on matters of public interest. A generated lifestyle scene that reads as an ordinary photograph sits inside that territory. If you use the icons, the display rules are specific: the label must be clearly perceivable at first exposure, embedded in the content itself rather than only in the platform interface, and it has to survive resharing and downloads. The Commission’s own user testing found the basic icon worked best when paired with a plain text label such as “modified”.

The human-oversight nuance from the draft phase made it into the final Code. AI-generated text published to inform the public gets lighter treatment when a human has reviewed it and taken editorial responsibility. Evidently artistic, creative, or satirical works only need disclosure that does not hamper their display. The provision is written for text and creative work. The underlying logic is that human oversight changes the compliance calculation. Regulators will bring that same frame to product imagery. Fully automated AI output is one thing. AI output that a human professional has reviewed, retouched, and signed off on is something else.

Penalties are real. Article 50 transparency violations sit under Article 99(4) of the Act. Fines reach €15 million or 3% of global annual turnover, whichever is higher. The headline €35 million / 7% tier applies only to Article 5 prohibited-use violations, not to disclosure breaches. German market surveillance authorities have already opened investigations against major eCommerce platforms over AI-generated product imagery. Enforcement is under way.

United States

This is the hybrid case. Federal posture leans accuracy. State laws push transparency, and the two are in active tension. There is no federal AI content disclosure law. Regulation comes from two directions: FTC guidance at the federal level, and a growing patchwork of state statutes.

The FTC’s March 2025 staff guidance set three principles for AI in advertising. If AI tools generate or substantially modify ad content, the seller should disclose the AI involvement. Any claim made by AI-generated content has to be truthful and substantiated regardless of how the content was made. And if AI produces something that looks like a personal endorsement or testimonial, the AI involvement must be clearly disclosed. Section 5 civil penalties reach $53,088 per violation under the 2025 inflation-adjusted schedule, which the FTC has carried into 2026 after OMB cancelled this year’s automatic adjustment.

In December 2025, an executive order directed the Attorney General to challenge state AI laws considered to obstruct national competitiveness. How that resolves will shape compliance through 2027.

Key state laws to track:

California AI Transparency Act. Requires latent and manifest disclosures in AI-generated content. In force since August 2, 2026, aligned with the EU.
New York Synthetic Performer Disclosure Law, General Business Law section 396-b. In force since June 9, 2026. Requires a conspicuous disclosure when an advertisement features a synthetic performer, meaning a digitally created asset that creates the impression of a human performer, is not recognizable as an identifiable real person, and is not a digital replica. The trigger is the use of the synthetic performer, not the use of AI as such. Civil penalties are $1,000 for a first violation and $5,000 for each subsequent one.
New York Fashion Workers Act. Requires clear, conspicuous, prior written consent before creating or using a model’s digital replica. Consent must specify scope, purpose, duration and compensation. The law leaves out routine retouching. Applies to clients and management companies, with civil penalties.
Georgia and Massachusetts. Bills in progress that would require disclosure whenever a seller uses AI-generated content in advertising or commerce.

Both New York statutes bite hardest for apparel brands putting AI models in the frame. Our guide to AI photoshoots at catalog scale covers which images trigger them and what your model releases need to say.

For any seller whose listings are visible nationwide, the practical move is to comply with the strictest active rule. Today that means California and New York. Sellers who would rather not track all five state regimes in-house can offload most of the labelling and record-keeping side to a managed service like Vision by Lucid Modules, which embeds disclosure metadata at generation time.

China

Pure transparency, the strictest version anywhere. China’s Measures for Labeling AI-Generated Content took effect on September 1, 2025, and they are the most detailed AI image labeling framework in force.

Every piece of AI-generated content distributed on a Chinese platform must carry two labels. The explicit label is a visible indicator inside the file itself: a watermark, caption, disclaimer, or audio cue. It tells the viewer AI made the content. The implicit label is a technical identifier in the file’s metadata, with encrypted fields and content IDs that let platforms detect and trace AI content. A label in the interface only fails the rule. The marker has to sit inside the file.

Scope is broad: WeChat, Douyin, Weibo, Xiaohongshu, plus Tmall, JD.com, and Taobao. Selling on any of those platforms means every AI-generated image has to comply.

An amended Cybersecurity Law has been enforceable since January 1, 2026. It layers on AI security review requirements and data localization, and it removes the previous warning-shot approach. First-time violations can now draw immediate fines.

United Kingdom

This is an accuracy story. No transparency mandate, but the older “does the image honestly show the product” rule still binds. The UK is the outlier. As of May 2026, there is no UK law requiring AI-generated content to be labelled. The government has acknowledged the case for clearer labelling but cited technical challenges and committed to no timeline. A long-expected AI bill did not appear in 2025. In March 2026, the government published a report on copyright and AI. It proposes work with industry on labelling best practice and stops short of any mandate.

UK consumer protection law still applies. Misleading commercial practices regulations require product images to accurately represent what the customer receives, and the method of creation is irrelevant. Accuracy is the active obligation. Transparency is not, at least not yet.

South Korea

Transparency, with sector-specific accuracy enforcement bolted on. South Korea’s Framework Act on Artificial Intelligence took effect in January 2026 and introduced specific transparency duties for AI-generated content. Anyone who creates or edits AI-generated content and posts it online must indicate clearly that they used an AI tool. Removing or damaging an AI label is itself a violation. Platforms have to provide labeling tools and notify users of their obligations, and failure to comply produces fines for both the platform and the user.

The Korean government has paired this with stricter monitoring of AI-driven advertising in food, pharmaceuticals, cosmetics, and private education. Those are the sectors where false claims do the most consumer harm.

Marketplace policies

Mostly accuracy stories, with Amazon’s disclosure rule the one transparency exception. Government rules set the floor. Marketplaces add their own policies on top, and those policies are usually what gets a listing pulled.

Amazon

The Amazon AI image policy is the clearest of the big platforms. Since July 22, 2026 it also names the exact tag. Amazon allows AI-assisted edits like background replacement, color correction, and lighting adjustments. It prohibits AI generation that misrepresents the physical product: wrong colors, missing features, fake scale. The July 2026 rule adds a disclosure duty for one category. Images and videos that contain photorealistic AI-generated people need a label before they reach a listing or A+ content. The label is metadata. Add the keyword contains-synthetic-performer to the dc:subject XMP field with an IPTC-compatible metadata editor. Amazon then shows a notice on the listing that the image includes AI-generated people. Images of real people stay out of scope even when AI edited them. The rule follows the New York synthetic performer law that took effect on June 9, 2026.

Amazon has deployed machine learning detection for AI-generated imagery, and trade press reporting on the 2026 seller updates describes high-confidence flagging at scale. Amazon is expected to keep expanding listing suppression for non-compliant images through 2026. The working rule for sellers: use AI freely for backgrounds, lighting, and scene composition, keep the product itself accurate, and tag any AI-generated person before upload.

Shopify

Shopify has no AI-specific restrictions on product photography. Their image recommendations focus on file size and resolution, with 2048×2048 at a 1:1 aspect ratio called out as the preferred format. FTC deception rules still apply to US-facing stores, so the disclosure logic holds even when Shopify itself stays quiet.

Etsy

Etsy’s accurate-representation policy is the constraint to watch. For handmade or one-of-a-kind items, AI-generated images that do not show the real product can conflict with that policy. For mass-produced goods, Etsy generally accepts AI product photography. A January 2026 seller policy update reinforced the representation requirement without introducing AI-specific rules.

eBay and Walmart Marketplace

Both accept AI-generated product photos as long as they meet the technical specs for dimensions, backgrounds, and formats. Neither has added AI-specific disclosure rules. Quality and accuracy are still the active levers.

Allegro

Poland’s largest marketplace has no explicit AI photo policy. Main product images need a white background with no extra graphics, in line with the long-standing rules. Allegro has begun adding API flags for catalog data co-created with AI. The platform looks like it is building the transparency infrastructure that a later release could make mandatory.

What to do now

Three things are worth doing this quarter, regardless of where you sell.

Get the product accuracy question settled first. Every regulation and every marketplace policy agrees on this point, and it is the only requirement every market already enforces. If an image misrepresents the physical product, no amount of disclosure language fixes it.

Build disclosure into the listing template so it stops being an editorial decision. One line in the product description, “Product images enhanced using AI”, satisfies most jurisdictions. It is low-cost insurance against rules that are still tightening. For EU-facing listings, the official EU icons handle the visible half of the disclosure. The Commission’s own testing recommends pairing the icon with a short text label. On Amazon the label lives in the file: tag any image with an AI-generated person as contains-synthetic-performer before upload. Adding it once at the template level is faster than adding it 4,000 times now that the EU deadline has landed.

Keep originals and tag what is AI. Keep the source photos for every SKU, and maintain an internal flag that records which images were AI-generated or AI-modified. When a marketplace runs detection or a regulator asks for evidence, that record is the difference between a quick response and a delisting.

Dates worth keeping on the calendar:

In force since July 22, 2026: Amazon’s metadata tag for AI-generated people in listing images.
In force since August 2, 2026: EU AI Act Article 50 transparency obligations, plus the California AI Transparency Act.
December 2, 2026: end of the Article 50(2) marking grace period for AI systems placed on the market before August 2, 2026.
Through 2026: Amazon’s expanding ML detection and listing suppression for non-compliant AI images.
2026-2027: US federal preemption challenges to state AI laws.
Ongoing: Chinese enforcement under the September 2025 Measures and the amended Cybersecurity Law.

Who ends up holding the record

Disclosure, metadata embedding, and record-keeping are doable by hand. The work is real, and it grows with every SKU. Self-service AI tools push that work onto the seller. Managed services that include human review and compliance handling absorb most of it.

Vision by Lucid Modules handles the metadata, the labelling, and the record of which images are AI-generated. That covers the seller side of the dual obligation before the file reaches a listing. Generate compliant product images here.

References

European Union:

United States:

United Kingdom:

China:

South Korea:

Marketplaces:

Snapshot as of September 2026. Rules are moving. Check the primary source before any compliance decision.