# Which Images in Your Store Need an AI Label? · Easy AI Act Image Labels > Which AI images to label in your store and which not: table by image type, the potential-to-deceive test, rules for ads, social media, marketplaces. URL: https://easyaiactlabels.com/en/guides/which-images-need-a-label/ [Guides](/en/guides/) /Obligation # Which Images in Your Store Need an AI Label? Last updated: 16 September 2026 Responsible: Martin Jäger 9 min read Short answer A label is needed for images that look real although they are not: fully AI-generated product photos, AI models with a real product, photorealistic lifestyle scenes, and AI depictions of real people. No label is needed for a real product photo on an AI background as long as the image does not mislead about the product, for standard editing such as cutouts or color correction, or for visibly stylized illustrations, logos, and text graphics. Under the Commission’s guidelines of July 20, 2026, the test is the potential to deceive, not the technology; for photorealistic borderline cases, label when in doubt. Since August 2, 2026, Article 50 of the EU AI Act (Regulation (EU) 2024/1689) also applies to online stores. It does not require a label on every image that AI has touched, only the visible disclosure of images that look real but are not. This guide sorts the typical image types of a Shopify store into label, no label, or borderline case, following the European Commission’s guidelines of July 20, 2026 (C(2026) 5054 final) and the Code of Practice of June 10, 2026. ## Do I have to label AI images at all as a merchant? If you use a finished AI tool (Tinker, Magnific, Canva, ChatGPT, the Shopify image editor), you are a deployer under Article 3(4). For images, deployers are bound only by Article 50(4): deepfakes must be visibly disclosed as artificially generated or manipulated, under Article 50(5) in a clear and distinguishable manner, at the latest when someone first sees the image. The machine-readable marking in the file (C2PA, IPTC) under Article 50(2) is the generator provider’s responsibility, not yours. For generators already on the market before August 2, 2026, that deadline runs until December 2, 2026 (Article 111(4) as amended by Regulation (EU) 2026/1744); stores get no grace period. Under the guidelines (para. 113), an image is a deepfake if it appreciably resembles a person, an object (consumer goods included), a place, or an event, could plausibly exist, and would falsely pass as authentic. The test is objective, needs no intent to deceive, and looks at the expected audience. See the [guides overview](/en/guides/). ## Which image types need a label and which do not? The classification follows the positive and negative lists in the guidelines (paras. 113 to 116); anything beyond them is flagged as interpretation or uncertain. Image type Label? Reason Fully AI-generated product photo or photorealistic rendering that flatters, alters, or upgrades the product Yes Positive list: can mislead about appearance, features, or use Fully AI-generated product photo that shows the product accurately Borderline, label when in doubt Not addressed by guidelines or example catalog (uncertain, as of September 1, 2026) AI image of a product that does not exist in this form (color, detail, features) Yes Positive list; also misleading (Section 5 UWG Germany, Section 2 UWG Austria), which no label cures AI model or virtual influencer with a real product Yes Para. 113: “persons” includes realistic AI avatars; teleshopping and influencer examples, no artistic exemption Photorealistic AI lifestyle or usage scene Usually yes Para. 113: realistic places, scenes, depictions of services AI image of a real, recognizable person Yes Positive list; in Germany also the right to one’s own image (Sections 22, 23 KUG) Real product photo in front of an AI-generated background No Negative list, unless the ad misleads about the product Retouched photo: color correction, denoising, upscaling, background replaced for clearly aesthetic purposes No Para. 116: standard editing with minor impact, case by case Background removed (classic cutout) No Not named in para. 116; standard editing by interpretation while the product stays unchanged Generative cutout that reinvents product edges or hidden parts Borderline, label when in doubt Not named in para. 116 (uncertain, as of September 1, 2026) Illustration, icon, cartoon, visibly artificial look No Negative list; para. 113: unrealistic depictions without potential to deceive Physically impossible content (flying objects, talking animals) No Negative list, also in advertising AI stock image from an image library Unclear Whether the licensee counts as the deployer is officially unsettled (uncertain, as of September 1, 2026); label when in doubt if the image sits next to a product, not if it is purely decorative Logo, favicon, pure text graphic (sale banner, typography tile) No No person, object, place, or event shown (interpretation); the guidelines list AI ad copy as a negative example ## Why does the potential to deceive count, not the technology? The Commission draws the line not at “AI yes or no” but at whether the image can mislead about the real product. Rule of thumb for the gray zone: if the AI changes how the product itself looks, the image needs a label, at least AI MODIFIED. If it only changes the surroundings of an unchanged product, standard editing under para. 116 usually applies. - **The artistic exemption is of little help in advertising.** Recognizably commercial content falls outside it (paras. 120 to 122). - **A label does not cure a misleading image.** Section 5 UWG covers misleading statements about essential characteristics of goods, a product image is such a statement, and Article 50(6) leaves unfair competition law untouched. If the image shows something other than what you deliver, the image itself is the problem. ## What do you do in a borderline case? For the fully AI-generated but accurate product image there is no official answer and no court decision in Germany, Austria, or Switzerland (uncertain, as of September 1, 2026). Three positions: - **AI service desk of the Austrian regulator RTR**, FAQ, accessed August 31, 2026: narrow reading, no obligation without a deepfake. - **Wettbewerbszentrale (German unfair-competition association)**, guidance version 2 of July 29, 2026: AI images are “to be labeled as such when in doubt”. - **Schutzverband gegen unlauteren Wettbewerb (its Austrian counterpart)**, position of February 4, 2026: label photorealistic advertising images in general, clearly surreal ones not. The latter two can issue warning letters (Abmahnungen) in their countries, and both take the broad reading. So: with photorealistic images, label when in doubt. The decision stays with you; document which images come from which tool. Annex 1 of the Code and the guidelines (paras. 58, 59) decide between the two worded icons with one question: was there a source photo? Starting point Icon No source photo, the image comes from the prompt AI GENERATED Real photo, substantially altered by AI AI MODIFIED Real photo, standard editing only no icon If in doubt between the two, choose AI GENERATED, the stronger disclosure. ## Which typical cases come up in Shopify stores? - **Tinker, usage scene with “Model = Yes”**: a photorealistic person who does not exist uses your product. Clear positive-list case, label mandatory. - **Magnific lifestyle shot from two to four reference photos**: photorealistic scene, product in an invented setting, often with a person. Label, usually AI MODIFIED, since a real photo was the source. - **Shopify image editor, “Generate background”** on an unchanged product photo: standard editing, usually no label. Not everything needs a label: in existing catalogs, standard editing under para. 116 is the normal case and needs no label; for photorealistic borderline cases, label when in doubt. ## Where in the store does the label have to appear? The obligation applies to each individual image and to every person who sees it (para. 143), not to the store as a whole, so every image location counts: product gallery and variants, collections, theme sections such as slideshow and image banner, blog and page images, email templates. Not enough: a footer note, the terms of service, or a separate subpage (para. 142); metadata in the file (para. 117: deployers cannot rely on the provider’s marking); a tooltip or alt text alone (Code, sub-measure 1.2.1(a): recognizable without user action). What works: a visible badge on the image, top right without overlapping elements (Code, sub-measure 1.2.2), or a caption clearly assigned to the image; the Code treats an overlay on the image, such as one from a theme app embed, as equivalent to the burned-in icon (sub-measure 1.2.1(c)). Alt text belongs alongside the visible label (Article 50(5), second sentence, refers to accessibility requirements); on its own it is not enough. The official EU AI icons are the Commission’s three English ones (“AI”, “AI GENERATED”, “AI MODIFIED”; page version of August 10, 2026); translated wording is not an official file. ## What applies to social media, marketplaces, and feeds? An overlay from a theme app embed does not leave the store with the image. Ads, social posts, marketplaces, and newsletters need a labeled derivative; the canonical file stays without a burned-in icon. The platform rules below follow the platforms’ own statements as of August 31, 2026; they change often, so check before every campaign: - **Google Ads and Merchant Center**: Google exempted AI labels from its ban on text overlays and watermarks (policy article of July 9, 2026). Merchant Center requires the IPTC DigitalSourceType value on generated feed images and forbids removing it; whether that value survives the path from Shopify to the Google channel has not been measured (both as of September 14, 2026). - **Meta, TikTok, Pinterest**: all three combine automatic detection (IPTC or C2PA signals) with self-declaration (Meta: label “AI info”); TikTok requires the label for realistic AI content. - **Amazon**: the main image must show the actual product; a burned-in badge there violates the image guidelines (no text, no logos). Per the Wettbewerbszentrale (guidance version 2 of July 29, 2026), under the AI Act you, not the platform, are responsible for your generated media and their labeling. ## Can software make the decision for you? Only in part. Many generators write a provenance marker (C2PA, IPTC DigitalSourceType) into the file on saving. A found marker is evidence of AI origin and, with the matching IPTC value, also separates fully generated from edited. A missing marker says nothing: - On September 1, 2026, 34 of 47 text-to-image models in one generator catalog (Magnific/Freepik) set a marker, 13 did not; the Shopify image editor wrote none. - Screenshots and format conversions on download delete every marker. - Copies delivered by the Shopify storefront still carried a marker in only 5 of 34 cases on September 1, 2026, and 0 of 3 on September 10, 2026, so check the marker on the original in the Files section of your admin. An app like Easy AI Act Image Labels reads such markers on the original and proposes labels. Sorting the unmarked images by the table above remains your decision; more in [Evidence or hint](/en/guides/evidence-or-hint/). ## Frequently asked questions ### Do I have to label AI images created before August 2, 2026? The regulation contains no grandfathering clause for individual images. In its FAQ on the AI Act (as of August 31, 2026), the European Commission states that content generated before August 2, 2026 does not have to be labeled retroactively, but recommends that deployers do so. That is an authority’s view, not a rule: if you touch the catalog anyway, label as you go, and in a dispute you would have to document when an image was created. ### Is a sentence in the footer or in the terms of service enough? No. The Commission’s guidelines (para. 142) name menu depth and terms of use as examples of information that is easily overlooked. The disclosure belongs on the individual image, as a visible badge or a clearly assigned caption, supplemented by the alt text. ### Is a note like “image for illustration purposes only” enough? No. That phrase says the image does not show the specific product, not that it was made by AI. Those are two different statements; if both apply, you need both. ### Does the label have to be in the customer’s language? The AI Act contains no language rule; the benchmark is that the audience understands the disclosure. The Code of Practice sets the English abbreviation “AI” as a sign recognized across Europe. In Germany and Austria English is not ruled out as such, though the Wettbewerbszentrale recommends German wording for audiences with little tech affinity; for France, caution applies because of the Loi Toubon (uncertain, as of September 1, 2026). ### Does the obligation also apply to AI text in product descriptions? Not under the guidelines. The second subparagraph of Article 50(4) covers only AI text that informs the public on matters of public interest, and the Commission names advertising and product copy as a negative example. A store blog on factual topics remains an open case (uncertain, as of September 1, 2026). ### Can an app tell me which of my images are AI images? Only those whose generator wrote a provenance marker into the file. A found marker is evidence, a missing marker is no statement: on September 1, 2026, 13 of 47 text-to-image models in one generator catalog set no marker, and screenshots delete every marker. Sorting by the table remains your decision. ## Sources 1. Regulation (EU) 2024/1689 (AI Act), Articles 3, 50, 111. [publications.europa.eu](http://publications.europa.eu/resource/celex/32024R1689) (accessed 31 August 2026) 2. Regulation (EU) 2026/1744 (Digital Omnibus on AI), Article 1 points 20 and 39. [publications.europa.eu](http://publications.europa.eu/resource/celex/32026R1744) (accessed 31 August 2026) 3. European Commission, Guidelines on the implementation of the transparency obligations under Article 50 (Annex to C(2026) 5054 final, July 20, 2026). [ec.europa.eu](https://ec.europa.eu/newsroom/dae/redirection/document/131215) (accessed 31 August 2026) 4. European Commission, landing page of the Article 50 guidelines. [digital-strategy.ec.europa.eu](https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems) (accessed 31 August 2026) 5. Code of Practice on Transparency of AI-generated Content (June 10, 2026). [ec.europa.eu](https://ec.europa.eu/newsroom/dae/redirection/document/129555) (accessed 1 September 2026) 6. European Commission, EU icons for labelling AI-generated content (page version August 10, 2026). [digital-strategy.ec.europa.eu](https://digital-strategy.ec.europa.eu/en/policies/eu-icons-labelling-ai-generated-content) (accessed 31 August 2026) 7. Wettbewerbszentrale, AI complaints office (KI-Beschwerdestelle). [wettbewerbszentrale.de](https://www.wettbewerbszentrale.de/beschwerdestelle/ki-beschwerdestelle/) (accessed 31 August 2026) 8. Google Merchant Center, image guidelines (IPTC DigitalSourceType on generated images). [support.google.com](https://support.google.com/merchants/answer/6324350) (accessed 14 September 2026) This article is general information, not legal advice. Whether an image needs a label is your decision; for your specific case, please ask a lawyer. Read next [03 · Technology Evidence or Hint: How Do You Recognize AI Images?](/en/guides/evidence-or-hint/) [04 · Technology Content Credentials and C2PA: What Is Inside an AI Image?](/en/guides/content-credentials-c2pa/) I find the AI evidence. You confirm with one click. [Coming soon to the Shopify App Store](/en/how-it-works/)