AI Product Images for Amazon: Tools, Rules, and EU AI Act Guide

AI-generated product images are fundamentally transforming the Amazon marketplace. While professional studio photography used to cost between 75 and 150 euros per image, AI tools now create high-quality lifestyle images and backgrounds for a fraction of the price. Since August 2, 2026, the transparency obligations of the EU AI Act apply on top, and Amazon has introduced its own labeling requirement for AI-generated people in parallel. Every seller should know both, even though the most common use case, a real product photo plus AI staging, still works without any labeling.

In this article, we cover what Amazon allows, what the EU AI Act really means for marketplace sellers, and how to use AI images to boost conversions without violating guidelines. You can download our complete EU AI Act guide for marketplace sellers as a free PDF.

Updated in August 2026: Amazon's new metadata requirement for photorealistic AI people, the final EU guidelines on Art. 50 of July 20, 2026, and the Digital Omnibus Regulation (EU) 2026/1744 have been incorporated.

Amazon's Rules for AI-Generated Product Images

Amazon clearly distinguishes between different image types and applies different standards to each:

Main Image: Real Product Photo Required

The main image must be a professional photo of the actual product. Amazon requires a pure white background (RGB 255, 255, 255), the product must fill at least 85% of the frame, and minimum resolution is 1,000 pixels (2,000 recommended for zoom). Purely AI-generated main images are not permitted. For full technical requirements, see our product photos guide.

Secondary Images: AI Is Allowed

For secondary images (positions 2-9), AI usage is much more flexible. Amazon permits:

• AI-generated lifestyle backgrounds: Product placed in natural settings (kitchen, living room, outdoor)
• AI-enhanced infographics: Product features visually highlighted with generated elements
• AI-assisted editing: Background removal, shadow optimization, color correction

Important: Even with AI-generated secondary images, the product itself must not be artificially altered. The representation must match the real product, as misleading images lead to "not as described" returns and increase the Order Defect Rate (ODR). If your ODR exceeds 1%, you lose the Buy Box for up to 60 days.

New Since Late July 2026: Amazon's Labeling Requirement for AI People

In late July 2026, Amazon introduced its first genuine labeling requirement for AI content. It does not cover all AI images, but one specific case: listing images and A+ content showing photorealistic, fully AI-generated people. Such images must be technically marked before upload. Specifically, Amazon requires sellers to use an IPTC-compatible metadata editor to write the keyword contains-synthetic-performer into the dc:subject (XMP) field of the image file. Amazon does not require a visible label in the image itself; the disclosure lives in the metadata.

According to Amazon, the following are exempt:

• Images without people (the classic product image case, even with an AI background)
• Real people, even if their appearance was edited with AI
• Non-photorealistic depictions (illustrations, comic style)
• Known characters from movies, TV, or games

Watch out with lifestyle scenes: AI-generated people in the background count too. If you let AI place a real product photo into a busy scene, check whether synthetic people were created along the way.

The rule stems from a US law: New York's General Business Law Section 396-b has required disclosure since June 9, 2026, whenever advertising shows photorealistic AI people (fines: USD 1,000 for the first violation, USD 5,000 for each subsequent one). According to the announcement, however, Amazon is implementing the tagging marketplace-wide, not just for US listings. Amazon's public documentation names no grace period, and non-compliant images can be removed or listings suppressed under Amazon's existing image policies. Whether previously uploaded legacy images must be re-tagged is something Amazon's documentation leaves open so far. Our recommendation: set the tag consistently on every image containing an AI person, regardless of marketplace. It costs nothing, takes seconds in tools like ExifTool or Photoshop, and covers the Amazon requirement.

Real Photo Base vs. Fully Synthetic Image: What Matters Legally

The single most important distinction in this whole topic is almost always overlooked in practice: it is not "AI image" or "no AI image," but a spectrum. And it is exactly this axis that decides whether you are on safe legal ground. For Amazon and other marketplaces, only one point on that spectrum is genuinely relevant.

The Spectrum from Technical Correction to Full Synthesis

• Pure image editing: Cutouts, exposure, color calibration, background removal. Legally entirely uncritical, no labeling, this has been standard practice for years even without AI.
• Real product photo as the base plus AI staging: You take a real cutout of your actual product and let the AI build only the environment, meaning lifestyle background, scene, shadows, plus USP overlays and infographic text. The product itself, the one the customer ultimately receives, stays unchanged. This is the marketplace standard case, the use case that actually matters in practice.
• AI alters the product itself: Retouching out scratches, deepening the color, making the material look more premium, glamorizing proportions. This is where it gets critical, independent of any labeling question.
• Fully synthetic image: The product itself is rendered or invented by the AI. Highest risk across every area of law.

Perpendicular to the Spectrum: The Photorealistic AI Person

Since summer 2026 there is a second axis that runs independently of the product: is there a photorealistic, fully AI-generated person in the image? Your product can be a real photo and the AI model next to it still triggers two obligations: the visible disclosure under Art. 50 of the EU AI Act (more below) and Amazon's metadata tag contains-synthetic-performer. To avoid both, photograph a real model and let AI build only the environment, or leave people out of the image.

Why the Real Photo Base Is the Safe Path

When your source image is a real photo of the actual product and the AI contributes only environment, staging, and text overlays, you are well positioned across all three relevant areas of law:

• Amazon policy: This is exactly the path Amazon explicitly intends. AI is permitted for background, lifestyle, and infographics; only artificial alteration of the product itself is prohibited. A photo base satisfies this by definition.
• EU AI Act Art. 50: The labeling obligation targets "deceptively realistic" synthetic content, meaning images that pretend to be a genuine photographic capture. A clearly staged advertising composite with a real product and USP text is a classic product representation, not a deceptive deepfake. Text overlays are classic graphic design anyway, not a "manipulation" of the image content in the sense of Art. 50. The line runs at the photorealistic AI person in the image, see above.
• Unfair competition law: As long as the color, shape, proportions, and functions of the real product are depicted correctly, the staging is unproblematic under competition law. It only becomes misleading when the AI glamorizes the product itself.

Practical takeaway: The tool does not decide the legal risk, what the AI touches in the image does. Touching background, scene, and overlays is safe; touching the product itself is risky; and the AI person is its own labeling case. This is exactly the principle our own tool themarketplacetools.com follows: it builds on an uploaded real product photo and stages it, rather than synthetically recreating the product.

AI Image Tools for Amazon Sellers

Amazon sellers can leverage two categories of AI image tools: Amazon's own tools built into Seller Central and the Advertising Console, and specialized third-party solutions.

Amazon's Own AI Image Tools

Amazon introduced its own AI image generation tools for sellers and advertisers in 2024 and has continuously expanded them:

Image Generator in the Advertising Console

The Amazon Image Generator creates lifestyle backgrounds for ads directly within the Ad Console. Sellers upload a product image with removed background and describe the desired scene via text prompt (e.g., "product on a wooden table in a modern kitchen, natural light"). Amazon reports up to 40% higher CTR for ads with lifestyle context compared to product photos on white backgrounds (Source: Amazon Advertising).

AI Creative Studio (Canvas)

In early 2026, Amazon launched the AI Creative Studio (internally "Canvas"), which goes beyond background generation. It enables the creation of complete ad creatives including text overlays, badges, and call-to-actions. The tool is free for all sellers with active advertising campaigns.

Amazon Video Generator

Complementing the image generator, the Amazon Video Generator creates short video clips from product images for Sponsored Brands Video campaigns. This tool is also free and integrated directly into the Advertising Console.

Third-Party Tools

Beyond Amazon's own tools, a broad ecosystem of specialized AI image solutions has emerged:

• themarketplacetools.com: Our own AI platform with two relevant features: "All-in-One Listing" analyzes competitor reviews to generate USP-based product images that address real customer pain points. "Pimp My Image" enables logo adjustment, recoloring, and translation of existing images for international markets. Read more in our article Why we built our own Amazon tool.
• Midjourney / DALL-E / Stable Diffusion: General-purpose image AIs suited for lifestyle scenes and mood images. Advantage: maximum creative freedom. Disadvantage: product details are frequently distorted, which can lead to compliance issues.
• Specialized e-commerce tools: Platforms like Photoroom, Claid.ai, and Pebblely that are specifically optimized for product photography and deliver more consistent results than generic AI models.

Conversion Impact: What AI Images Actually Deliver

The data on conversion impact of AI-generated images is becoming increasingly robust:

• BigCommerce study (2025): AI-optimized product images led to a median conversion increase of 49% compared to non-optimized images (Source: BigCommerce Research)
• Amazon Advertising data: Ads with AI-generated lifestyle contexts showed up to 40% higher CTR than product images on white backgrounds
• 3D models vs. 360-degree images: Amazon discontinued 360-degree images on January 20, 2025, replacing them with 3D models. Sellers report 20% fewer returns for products with 3D views
• Cornell study: Consumers cannot reliably distinguish AI-generated images from studio photos; the detection rate was 51.3%, essentially random chance

Cost Comparison: AI vs. Studio

The cost advantage is substantial:

• Professional studio photography: EUR 75-150 per image (excluding styling, location, model)
• AI-generated lifestyle images: EUR 0.05-2.00 per image (tool subscription)
• Amazon's own AI tools: Free for active advertisers

For a typical Amazon listing with 7 images, that means: studio costs of EUR 525-1,050 vs. AI costs of under EUR 15. With an international rollout across 8 EU markets, this advantage multiplies.

EU AI Act: Timeline and Obligations in Detail

The EU AI Act (Regulation 2024/1689) entered into force on August 1, 2024. Obligations are phased in:

• February 2, 2025: Ban on unacceptable AI practices + Article 4 AI literacy obligation (already active!)
• August 2, 2025: Article 99 sanctions provisions become applicable
• July 20, 2026: The European Commission publishes the final guidelines on the Art. 50 transparency obligations (C(2026) 5054) and confirms the Code of Practice on Transparency of AI-Generated Content as an adequate compliance instrument
• August 2, 2026: Article 50 transparency obligations for AI-generated content, in force since this date
• December 2, 2026: Watermarking obligation for generative AI systems already on the market before August 2, 2026. This transition period is part of the Digital Omnibus Regulation (EU) 2026/1744, published on July 24, 2026 and in force since July 27, 2026

1. Risk Categories: Where AI Product Images Fall

The EU AI Act classifies AI systems into four risk levels:

• Unacceptable risk: Prohibited (e.g., social scoring)
• High risk: Strict regulation (e.g., AI in hiring, credit decisions)
• Limited risk: Transparency requirements for generative AI tools
• Minimal risk: No regulation (e.g., spam filters)

AI image generators for product photos fall under "limited risk": transparency obligations yes, approval procedures no. You can check your own AI classification using the official EU AI Act Compliance Checker.

2. Provider vs. Deployer: Who Bears Which Obligation?

The EU AI Act distinguishes clearly between two roles in Article 3:

• Provider: Anyone developing or placing an AI system on the market under their own name. In the Amazon context: OpenAI (DALL-E), Midjourney, Amazon itself (for its own AI tools)
• Deployer: Anyone using an AI system under their own authority. In the Amazon context: you as a seller or vendor

Important: As a deployer, you are not exempt from obligations just because Amazon acts as the platform. Both sellers and vendors bear independent regulatory responsibility for their AI usage in listings.

3. Article 50: The Five Transparency Scenarios

Article 50 defines five scenarios that trigger transparency obligations:

1. Interactive AI mimicking human behavior (e.g., chatbots)
2. Synthetic images, audio, or video that "falsely appear authentic"
3. Emotion recognition or biometric categorization
4. Deceptive deepfakes
5. Text content aimed at shaping public opinion

For sellers, scenario 2 is most relevant: AI-generated lifestyle images that look like real photographs. The specific obligations:

• Provider obligation: AI providers must label all outputs in a machine-readable format (cf. Recital 133)
• Deployer obligation: You as a seller may not remove provider markings and must add visible notices for "deceptively realistic" synthetic images (e.g., "AI-generated"). The final guidelines specify: the notice must be perceivable right where the content is perceived. Pure metadata or a note in the bullet points and product description do not satisfy this obligation; the safe route is a label in the image itself. On Amazon, that means in practice: on the gallery images, since text and logos do not belong on the main image anyway
• Exception: Assistive AI functions that do not substantially alter the input data or their meaning are exempt from labeling. A real product photo that AI merely stages regularly falls under this exception according to the final guidelines. Photorealistic, fully AI-generated people, by contrast, count as synthetic content subject to labeling

Final guidelines and legacy images (July 2026): On July 20, 2026, the European Commission published the final version of its guidelines on Art. 50 (C(2026) 5054) and confirmed the Code of Practice on Transparency of AI-Generated Content as adequate. The line is now considerably clearer than in spring. Important for existing catalogs: images published before August 2, 2026 do not have to be labeled retroactively, according to the Commission. Voluntary re-labeling is recommended where possible without disproportionate effort, if only to preempt accusations of misleading practices.

Bottom line for Amazon sellers: Product images built on a real photo base with AI staging are not subject to the labeling requirement under the final guidelines, because they are not "deceptively realistic" synthetic captures. You need to act in two cases: photorealistic AI people (visible label in the image plus Amazon's metadata tag) and fully synthetic images. We recommend documenting AI usage internally and adding a visible notice when in doubt. Whether an AI label costs conversion is something you should test rather than guess: early research suggests labels can cost trust especially on higher-priced products, so the labeled variant belongs in your A/B test.

4. Penalties for Non-Compliance (Article 99)

Sanctions are linked to global annual turnover and applicable since August 2, 2025:

• Prohibited AI practices (Art. 5): Up to EUR 35 million or 7% of worldwide annual turnover, whichever is higher
• Violations of other obligations, including the transparency obligations under Art. 50: Up to EUR 15 million or 3% of worldwide annual turnover
• False or misleading information to authorities: Up to EUR 7.5 million or 1% of worldwide annual turnover

Missing or incorrect labeling of AI-generated images therefore falls into the middle category, up to EUR 15 million or 3% of turnover. For SMEs, whichever of the two values is lower applies.

For SMEs: The EU AI Act explicitly provides for proportionate reductions. Micro and small enterprises (under 50 employees, under EUR 10M turnover) receive priority access to regulatory sandboxes and simplified documentation templates.

5. Amazon Is Already Implementing: Its Own Labeling Mechanisms

That Amazon as a platform would introduce its own AI transparency mechanisms was foreseeable, similar to the existing GPSR compliance dashboard. What is remarkable is the trigger: the new metadata requirement for AI people (see above) does not come from the EU AI Act but from a New York law, yet Amazon is rolling it out marketplace-wide. Amazon's own AI tools ("Enhance My Listing" is already used by over 400,000 sellers globally) will presumably integrate the machine-readable provider marking automatically. For images from third-party tools, you as the seller are responsible, both for Amazon's metadata tag and for the visible disclosure under Art. 50.

6. Article 4: AI Literacy Obligation

An often-overlooked aspect of the EU AI Act is Article 4: the AI literacy obligation. It has been in effect since February 2, 2025 and applies to all companies using AI systems, regardless of size or usage frequency.

No formal training is mandated, but documentation is required. Specifically, sellers must be able to demonstrate:

• Internal training measures: Which employees were trained on which AI tools?
• AI usage policies: Documented internal rules for using AI-generated content
• Approval processes: Who reviews and approves AI-generated product images before upload?
• Tool documentation: Which AI tools are used, which prompts are employed, which content is created?
• Risk assessment: Awareness of risks (distortions, hallucinations, copyright questions regarding training data)

This is not a theoretical obligation: in case of complaints or legal challenges, missing documentation can be treated as a violation. Verification of commercial usage rights and the origin of training data for AI models used should also be documented.

Our complete guide with all risk levels, obligations, and a seller checklist is available as a free PDF download.

Current Debate: Criticism of the AI Labeling Obligation

Around the time the labeling obligation took effect on August 2, 2026, criticism of its actual design grew loud. In a widely noted commentary from July 14, 2026, the German retail association Händlerbund raises a point that is highly relevant for Amazon sellers in practice: the obligation attaches to the tool, not to the result.

Tool Instead of Result: The Flaw in the Rule

The Händlerbund's example nails it: if you remove your product photo's background by hand and drop in an alpine panorama, no labeling is required. If you use an AI generator for the exact same background, the obligation kicks in, even though the product itself is real and unchanged in both cases. What decides is whether AI was involved, not whether the image deceives anyone. This is precisely the distinction we described above as the one that actually matters: the tool does not determine the risk, what is changed in the image does.

The consequence of this logic: an honest shop showing a real product correctly must label an AI composite and may thereby look more suspicious than a dishonest seller who simply posts a deceptive image without any label. The label tends to hit the wrong parties. The exceptions in Art. 50(4) (obviously artistic, creative, or satirical works as well as law enforcement) offer little help in everyday marketplace practice.

What This Means for You as a Seller

The criticism does not change the fact that the obligation has applied since August 2, 2026. But it confirms our core recommendation: anyone working on a real photo base and letting AI build only the environment, staging, and overlays has no deception problem, because the real product stays depicted correctly. And for any genuine discrepancy between image and goods, two existing areas of law already apply today, entirely without an AI label: unfair competition law (UWG) and warranty law.

Beyond the AI Act: Unfair Competition, Warranty and Copyright

Most coverage of this topic talks only about the EU AI Act. But for Amazon sellers, three other areas of law are often more relevant in practice, because they did not start applying in August 2026, they have applied all along: unfair competition law, warranty law, and copyright.

Unfair Competition Law: Misleading Advertising Is Actionable Immediately

German unfair competition law (UWG) applies independently of the AI Act and without a transition period. If an AI image depicts the product as nicer, larger, shinier, or more vivid than it really is, that can be a misleading commercial practice, just like excessively retouched conventional advertising photography. The crucial point: it is not only authorities who can act on this, but above all competitors and enforcement associations. On a marketplace as transparent as Amazon, where your competition watches your listings daily anyway, that is a real and fast risk. Several legal commentators expect a genuine wave of cease-and-desist letters now that AI labeling has been in effect since August 2026.

Here too, the real photo base is the protection: anyone who depicts the actual product correctly and only stages the environment has not represented anything misleading. It only becomes critical when AI glamorizes the product itself.

Warranty Law: If the Product Deviates from the Image, It Is a Defect

The second area of law that applies today and entirely without an AI label is warranty law. A product image is part of the agreed condition of the goods: if the delivered item deviates noticeably in color, shape, size, or function from the depiction, there is a material defect. The buyer then has the usual remedies, from subsequent performance through price reduction and withdrawal to damages. Just as with unfair competition law, it is completely irrelevant whether the image was created with Photoshop, with AI, or conventionally in a studio. All that matters is whether the real product delivers what the image promises.

This is the real reason the tool orientation of the AI labeling obligation often runs into the void in practice: a noticeable discrepancy between image and goods has long been sanctioned, regardless of how the image was created. Anyone working on a real photo base is on the safe side here too, because the delivered product matches exactly what was photographed.

Copyright: Fully AI-Generated Images Are Often Unprotected

One aspect that is easily overlooked: fully AI-generated images generally enjoy no copyright protection under current German case law. The Munich Local Court ruled on February 13, 2026 (case no. 142 C 9786/25) regarding AI-generated logos that the defining personal creativity is absent when the creative design is essentially left to the rules of the AI. Applied to product images, this means a fully synthetic image could in theory be copied by a competitor without you being able to invoke copyright.

Here as well, the real photo base plays to its advantage. A real product photo is protected as a photographic work, and that protection is retained when the photo forms the basis of the staged image. Anyone working purely synthetically gives up that protection.

How Do Other Platforms Handle AI Labeling?

If you sell beyond Amazon, you will encounter very different rules. There is no universal obligation to label all AI images anywhere, but the triggers differ considerably:

• TikTok Shop: Requires a visible AI label for AI-generated product imagery, explicitly including AI models and fully synthetic lifestyle scenes. Without the label, you risk reduced distribution up to takedowns. Light AI editing such as color correction or background removal counts as normal post-production.
• Etsy: Since January 14, 2026, requires disclosure when generative AI created or altered the item being sold (checkbox in the listing plus "Designed by" instead of "Made by"). Pure AI editing of a photo of a real product does not require disclosure.
• Shopify and eBay: No AI label mandate of their own. General rules against misleading representation apply, plus Art. 50 of the EU AI Act when selling into the EU.

The pattern across platforms: labeling is becoming the norm, but each with a different trigger. Amazon looks at AI people, TikTok Shop at the AI image as a whole, Etsy at AI touching the product, the EU at the risk of deception. If you list internationally, the safest route is to follow the strictest rule in your set.

Best Practices: Using AI Images Correctly

Based on our work with over 15 brands, we recommend the following strategy for AI image deployment:

The Optimal 7-Image Strategy with AI

1. Main image (Position 1): Professional studio photo, no AI. White background, 2,000+ pixels, 85%+ fill
2. Lifestyle image (Position 2): AI-generated context. Product in natural environment, ideal use case, without photorealistic AI people
3. Infographic (Position 3): AI-assisted creation. Key features with icons and short text
4. Size comparison (Position 4): Product alongside everyday objects. AI can generate the background and scene
5. Detail shot (Position 5): Real photo of materials, textures, closures. Authenticity matters here
6. Usage image (Position 6): AI lifestyle scene showing the product in use. Careful: as soon as a photorealistic AI person uses the product, the image becomes subject to labeling (visible label plus Amazon's metadata tag). Safer route: photograph a real model and let AI build only the environment
7. USP/Trust image (Position 7): Certifications, awards, comparison with competitors. AI-assisted infographic

Common Mistakes to Avoid

• Product distortion: Generic AI models distort details. Use specialized e-commerce tools or upload the real product photo as reference
• Unrealistic representation: AI images that show the product larger, shinier, or more colorful than reality lead to returns. Amazon's return rate already sits at 5-15%
• Lack of consistency: All 7 images must visually cohere. Consistent style, color palette, and tonality are essential
• Forgetting mobile: 59.52% of Amazon traffic comes from mobile devices. AI images must work on small screens too

Checklist: AI Images for Amazon 2026

• Main image: Real studio photo (no AI)
• Secondary images: AI for lifestyle contexts and infographics permitted
• Product representation: Must match the actual product
• AI people: Label photorealistic AI models visibly in the image and tag them with contains-synthetic-performer (dc:subject/XMP) before upload
• EU AI Act Art. 50: Transparency obligations in force since August 2, 2026, identify the labeling cases in your image catalog
• Legacy images: Images published before August 2, 2026 need no retroactive labeling, voluntary review recommended
• Documentation: Document AI usage internally (which tool, which image)
• AI literacy: Ensure team training on AI tools in use
• Mobile check: Test all images on mobile devices
• A/B testing: Test AI vs. studio images, measure conversion

Conclusion: Leveraging AI Images as a Competitive Advantage

AI-generated product images are no longer a future trend but a standard tool for successful Amazon sellers. The cost advantage is enormous, the conversion impact proven, and the legal framework has become considerably clearer with the final EU guidelines and Amazon's own labeling rule.

The key lies in strategic combination: real product photos for authenticity, AI-generated images for context and emotion. Those who have built internal processes, ensure AI literacy within their team, and know the labeling cases in their own image catalog hold a clear advantage now that the obligations apply.

Download our free EU AI Act guide to learn which obligations apply and when. Want to optimize your product images professionally? Our SEO team creates AI-optimized listings that rank and convert.

Sources

• EU AI Act, Article 50: Transparency Obligations
• EU AI Act, Article 4: AI Literacy Obligation
• EU AI Act, Article 99: Sanctions
• EU AI Act Compliance Checker
• European Commission: Final Guidelines on the Article 50 Transparency Obligations (July 20, 2026)
• EU Code of Practice on AI-Generated Content
• Digital Omnibus Regulation (EU) 2026/1744 Overview (Praxikon)
• Forbes: Amazon Requires Sellers To Label AI-Generated People In Listing Images (July 25, 2026)
• Händlerbund: How to correctly label AI images in Amazon listings (in German)
• Händlerbund: AI labeling in online retail, why the label hits the wrong parties (July 14, 2026, in German)
• Amalytix: EU AI Act 2025 - Obligations for Amazon Sellers and Vendors
• Amazon Advertising: AI Image Generation Tools
• Seller Labs: Amazon Product Image Requirements 2026
• Herbert Smith Freehills: Transparency Obligations Under the EU AI Act
• EU AI Act, Article 5: Prohibited AI Practices