You only have to show your own label in your store if the image is a deep fake (Article 50(4) AI Act). For standard editing, a purely decorative AI background behind a real product, or images created before 2 August 2026, the European Commission's guidelines do not require one (paras. 92, 116, 154), and a real photo without AI should not carry one at all.
Never remove AI providers' machine-readable markings (C2PA, IPTC, watermarks): the Code of Practice provides for keeping them, Google Merchant Center requires it, and China and India prohibit removal by law.
Key facts
- A retailer's labelling duty covers deep fakes only: AI images that look real and can mislead (Article 3(60), Article 50(4) AI Act).
- An AI label on an image without AI can itself mislead. If it is there by mistake, take it down.
- If an image does not require a label under the guidelines, you may leave it out. You may still show one voluntarily if AI was involved.
- Platform notices such as Meta's “AI info” can come from metadata written by editing software. Whether a label fits depends on what the AI did to the image, not on the metadata alone.
- ai-label.app deliberately offers no tool that removes AI markings or watermarks.
When you may leave out your own label
The obligation under Article 50(4) AI Act applies only to deep fakes. Under the European Commission's guidelines a label is not required in these cases:
| Case | Under the guidelines | Citation |
|---|---|---|
| Background removal, colour correction, cropping, sharpening | not required | paras. 92, 116 |
| Purely decorative AI background behind a real, unchanged product | not required | para. 116 |
| Distracting details removed (passers-by, dust) | not required | paras. 92, 116 |
| Clearly unreal illustration | not required | para. 113 ii |
| Created with AI before 2 August 2026 | no retroactive label | para. 154 |
| Real photo without AI | not an AI image | Article 3(60) |
When in doubt, the guidelines and we recommend the label. Check your case with the labelling check; the result names the paragraph you can rely on.
Leaving out a label does not make a misleading image acceptable. If an image shows the product better than it is, it remains open to challenge under competition law, with or without a label (recital 137 AI Act, guidelines para. 129).
When a real photo is flagged as AI
Platforms such as Facebook and Instagram show notices like “AI info” when they find AI signals in a file, such as Content Credentials (C2PA) or the IPTC note. Editing programs also write such signals when you use generative features, for example generative fill. According to Meta, content that was only edited shows the notice in the menu rather than directly on the image.
- Check what is in the file: the AI image checker shows which signal triggers the notice.
- Check whether the notice is accurate: if a generative feature changed the image, the notice is correct even if the photo is essentially real.
- Use the platform's own process: platforms set their own rules for how notices appear. Each platform's rules are under platforms.
What you should never do
Markings an AI provider wrote into the file are how it complies with Article 50(2) AI Act. Removing them to hide the AI origin contradicts the Code of Practice, can mislead about origin and breaks platform rules:
- Google Merchant Center requires that embedded metadata such as the IPTC DigitalSourceType is not removed from AI images.
- China prohibits removing or falsifying AI labels (labelling measures since 1 September 2025), and so does India (IT rules since 20 February 2026).
- That is why ai-label.app has no tool that removes labels, watermarks or AI metadata. Our labelling tool carries existing metadata over into every new file.
Frequently asked questions
May I remove the AI label from my store again?
Yes, if the image is not a deep fake under the guidelines, for example standard editing or a purely decorative background (paras. 92, 116), or if no AI was involved at all. For deep fakes the label must stay as long as the image is shown.
How do I remove the “AI info” label on Instagram?
The notice comes from signals in the file or from your own disclosure. Use the AI image checker to see which signal is present and whether a generative feature actually changed the image. If it did, the notice is accurate. We do not recommend stripping metadata to avoid it.
Is there a tool to remove AI watermarks?
Not on ai-label.app. Removing AI providers' markings contradicts the European Commission's Code of Practice, breaks platform rules such as Google Merchant Center's, and is prohibited by law in China and India.
What if my supplier delivered images with an AI label?
The guidelines recommend keeping existing labels and informing customers, even if you are not the deployer yourself (paras. 16, 17).
Sources
All sources read in full. Paragraph numbers (para.) refer to the European Commission's guidelines on Article 50, C(2026) 5054.
- Commission Guidelines on the transparency obligations under Article 50 AI Act, C(2026) 5054 finalEuropäische Kommission · Commission · 20 Jul 2026
- Regulation (EU) 2024/1689 (AI Act)EUR-Lex · Law · 12 Jul 2024
- Code of Practice on Transparency of AI-Generated Content (final version)Europäische Kommission · Code of practice · 10 Jun 2026
- Meta: identifying and labelling AI contentMeta · Platform · 1 May 2026
- Meta: Ansatz zur Kennzeichnung KI-generierter InhalteMeta · Platform · 5 Apr 2024
- Google Merchant Center: AI-generated contentGoogle · Platform · 3 Oct 2026
- China: Measures for labelling AI-generated synthetic contentCyberspace Administration of China · Law · 14 Mar 2025
- India: IT (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules 2026Gazette of India · Law · 10 Feb 2026
Not legal advice. This content explains the law based on primary sources; it does not replace an assessment of your individual case.
AI-generated
