AI Disclosure for AI Images
AI-generated images should be labelled in the caption or metadata. The EU AI Act specifically calls out synthetic-media transparency.
Synthetic imagery is a core focus of transparency rules. The EU AI Act requires AI-generated or manipulated images to be clearly marked, and major platforms now read metadata (like C2PA content credentials) to auto-label AI images. Labelling in the visible caption, the alt text, and the file metadata gives you the strongest, most durable disclosure.
Real-world examples that need labelling
A fully AI-generated stock-style photo used on a website, an AI-generated portrait used as a profile picture presented as a real photo, a composited image where an AI tool added or removed a person or object from a real photograph, and AI-upscaled or AI-restyled photos shared as "art" all typically need a label. A minor colour correction or crop of an original photo does not.
Common misconceptions
- "Metadata alone is enough" — most viewers never inspect metadata, so a visible caption or on-image label is still necessary for real transparency.
- "Alt text disclosure is only for accessibility" — it also feeds machine-readable detection systems and search engines, making it a meaningful disclosure channel.
- "C2PA content credentials are mandatory everywhere" — they are a voluntary but increasingly adopted open standard, not yet a universal legal requirement outside specific jurisdictions.
- "Only fully AI-generated images need labelling" — substantially AI-edited real photographs (added/removed elements, face swaps) usually need disclosure too.
Practical guidance on labelling workflow
Build labelling into your export workflow: write the caption disclosure first, then write matching alt text, then check whether your generation or editing tool embeds C2PA content credentials by default (many mainstream AI image tools now do). If you strip metadata by re-saving or screenshotting, re-add a text disclosure since the embedded credential will be lost. For EU-facing publications, treat the visible label as the primary compliance mechanism, since it is the most verifiable and least dependent on tooling.
Current rules for AI-generated images
- Add a visible label in the caption or on the image where audiences will see it.
- Include AI disclosure in the alt text for accessibility and machine readability.
- Preserve or add content-credential metadata (C2PA) where your tools support it.
- The EU AI Act expects synthetic images to be clearly marked as artificially generated.
Example disclosures
This image was generated entirely by artificial intelligence.
AI-generated image: [describe the scene].
Generate a disclosure for AI-generated images
We’ve pre-filled the wizard for your use case. Answer a couple of quick questions and copy your statement.
Start freeFrequently asked questions
Do AI images legally need a label?
In the EU, yes — the AI Act requires synthetic or manipulated media to be clearly disclosed. Most major platforms also expect labelling.
Where should the AI image label go?
In the visible caption, the alt text, and the metadata where possible for maximum durability.
What is C2PA / content credentials?
An open metadata standard that records how an image was made, letting platforms detect and label AI-generated media automatically.
Do AI-edited photos count?
Substantial AI manipulation of a real photo should be disclosed; minor retouching generally does not.
Does cropping or resizing an AI image remove the need to disclose?
No — the image content is still AI-generated regardless of resizing, so the disclosure obligation stays the same.
Do I need to disclose AI-generated icons or graphics used in a UI?
Decorative UI assets are lower risk, but if presented as real photography or documentary imagery they should still be labelled.