Visible AI labels, in four sets
AI-generated, AI-manipulated, AI interaction and AI-assisted badges, each in brand light, brand dark and neutral, for the visible disclosures the EU AI Act asks of deployers. They are our designs, not an official EU mark.
AI-generated
Made by an AI system · ai-generated
AI-generatedAI-generatedAI-generatedAI-manipulated
An existing image, video or voice altered by AI (deepfake) · ai-manipulated
AI-manipulatedAI-manipulatedAI-manipulatedAI interaction
The reader is talking to an AI system · ai-interaction
AI interactionAI interactionAI interactionAI-assisted
Made by a person with AI help · ai-assisted
AI-assistedAI-assistedAI-assisted
Which badge for which disclosure
| Badge set | Use it for | Where the rule comes from (draft reading) |
|---|---|---|
| AI-generated | Images, video, audio or text an AI system produced | AI Act Art. 50(4): deployers disclose deepfakes and AI-generated text published to inform the public |
| AI-manipulated | An existing image, video or voice that AI altered so it could pass for an original (a deepfake) | AI Act Art. 50(4) |
| AI interaction | A chatbot or assistant a person talks to | AI Act Art. 50(1): providers make the interaction clear |
| AI-assisted | Work a person made with AI help | Not an Art. 50 category: a voluntary label for editorial policies |
The readings above are drafts for owner and legal review. The machine-readable side of Article 50(2), marking outputs in a detectable format, is what C2PA signing and watermarking do; a visible badge does not replace it.
The variants
Brand light and brand dark use the Original Pictures palette for sites in either mode. Neutral is black on white, for publishers whose design should not carry our colours. Every badge has a text label, not only an icon, and keeps its contrast at small sizes.
If an official mark appears
These badges do not replace an official label. If the EU or a member state publishes one for these disclosures, use it; the SDK's badge sets are versioned, so a new set can follow.
Original Pictures shows where content came from and whether it has changed since it was signed. It does not prove that a depicted event is true. Using it supports your work on the EU AI Act, the California AI Transparency Act and other AI labeling rules, and it is not by itself legal compliance with any of them. Nothing on this site is legal advice.
Place a badge. Or draw it over the image.
The label component logs that a label was offered or applied to your disclosure log. The badge() helper draws a label over an image in the browser and never changes the stored file.
Coming soon · Not published yet: use the REST API today.
<!-- Import once in your site's bundle:
import "@originalpictures/disclosure"; -->
<op-ai-label badge="ai-generated" variant="brand-dark"></op-ai-label>
<op-ai-label badge="ai-manipulated" variant="neutral"></op-ai-label>AI badge questions
Is this the official EU AI label?
No. These are Original Pictures designs for the visible disclosures the AI Act describes. The Act does not prescribe one icon. If an official mark is published, use that.
Does adding a badge change my file's Content Credentials?
No. The component and the badge() helper draw the label on the page. The stored file and its signed manifest are unchanged.
Is a visible badge enough for Article 50?
Not on its own. Article 50 also asks providers to mark outputs in a machine-readable way, which is what C2PA signing and watermarking cover. Whether your labels are enough is a question for your counsel.
Label what your system makes
Sign outputs with C2PA, and show a visible badge where a person sees them. Book a demo to plan both with your team.