You must label an image if you generated or changed it with AI and it works as a deep fake, meaning it shows people, products, places or events that look real but do not exist like that (Article 50(4), Article 3(60) AI Act). The check walks through these criteria.
How the check decides
It is based solely on the European Commission's guidelines on Article 50 (C(2026) 5054 of 20 July 2026). Three rules are built in:
- Conservative. “Unsure” always leads at least to “labelling recommended”, never to “not required”.
- The strictest result wins. If the AI did several things (say, background and person), the strictest part counts.
- No question about intent. Intent to deceive does not matter (para. 114). The yardstick is the widest audience, including children and people with little AI literacy (para. 115).
Possible results
| Result | Meaning |
|---|---|
| Labelling required | A deep fake under the criteria of the guidelines. |
| Lighter form | A deep fake in an evidently artistic work, such as an art print you sell (Article 50(4), third sentence). |
| Labelling recommended | The guidelines leave the case open, or an answer was unsure. |
| Third-party image | You are probably not the deployer for this image (paras. 14, 16); a label is still recommended. |
| Legacy image | Created with AI before 2 August 2026 (para. 154): no retroactive labelling required, but encouraged. |
| Not required under the guidelines | Standard editing or an insignificant change (paras. 92, 116). |
Typical cases with a result are in the case catalogue, such as AI model, AI background or background removal.
Why you decide
The guidelines are not binding; only the Court of Justice of the EU can interpret the law bindingly (para. 5). The check therefore does not legally assess your case; it maps your answers to the Commission's criteria. The Shopify app runs the same logic (tree.v1) and records every decision with its reasoning.
Frequently asked questions
Do I have to label every AI image?
No. The obligation under Article 50(4) covers deep fakes, meaning AI content that looks real and can mislead. Standard editing and purely decorative backgrounds behind a real product are generally not covered under the guidelines (paras. 92, 116). A voluntary label is never wrong, as long as AI was involved.
Does this apply to manufacturer images?
If you merely pass on third-party images without deciding on the use of AI, you are generally not the deployer under the guidelines (paras. 14, 16). The Commission still recommends keeping existing labels and informing customers (paras. 16, 17).
What about images I created before August 2026?
Deep fakes from before 2 August 2026 do not need to be labelled retroactively, though the guidelines encourage it (para. 154). What counts is the day of the AI editing, not the day of upload.
Are my answers stored?
No, the check runs in your browser. If you copy the result link, the answers are in the address after the # sign. That part is not sent to us.
Sources
Para. refers to the paragraph number in 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
- Regulation (EU) 2026/1744 (Digital Omnibus on AI)EUR-Lex · Law · 24 Jul 2026
