Metaesthetic AI

Which AI tools and platforms label AI-generated images?

Generators mark their images in two ways: C2PA Content Credentials, signed metadata that records origin and edits, and invisible watermarks such as Google's SynthID, which Google says is in every image generated with its Nano Banana models in the Gemini API. Platforms then read those signals to label posts: YouTube applies an AI label to content with C2PA metadata, TikTok reads Content Credentials to auto-label AI content, and Meta shows an "AI info" label when it detects industry-standard indicators or the poster discloses AI use. A missing mark proves nothing, because editing, converting or sharing a file can strip its metadata.

Who marks or labels AI images, and how?

Each row comes from the company's own documentation or announcement.

Company What it adds or reads What its page says
OpenAI C2PA Content Credentials (images) and SynthID watermarks (images and audio) as its provenance signals Its Content Provenance API and openai.com/verify check images for C2PA Content Credentials and SynthID, and audio for SynthID. It does not detect content made by other companies' models, so a not_detected result does not rule out AI
Google SynthID invisible watermark All images generated with the Nano Banana models in the Gemini API include a SynthID watermark. Gemini, Search and Chrome can check an uploaded file for one, and the SynthID Detector checks content from Google and partners including OpenAI, NVIDIA and Kakao
Adobe C2PA Content Credentials In March 2024 Adobe said Content Credentials are attached to Firefly generations and to assets created or modified with Firefly-powered features in Express
Meta Visible "Imagined with AI" marker, invisible watermark and IPTC metadata on Meta AI images; reads C2PA and IPTC Meta labels images from other companies' tools when it detects the "AI generated" information in the C2PA and IPTC standards. Since July 2024 the label reads "AI info"; since September 2024, content only edited with AI shows it in the post's menu
TikTok Reads and attaches C2PA Content Credentials In May 2024 TikTok began auto-labeling AI content uploaded from other platforms by reading Content Credentials, and said it would attach Content Credentials to TikTok content, calling itself the first video sharing platform to do so
YouTube AI disclosure labels YouTube may apply an AI label automatically to content made with its own generative AI tools, content that contains C2PA metadata, and content its systems detect as AI generated or altered. Creators cannot remove a label that came from C2PA metadata

What is C2PA?

The Coalition for Content Provenance and Authenticity (C2PA) publishes an open technical standard, called Content Credentials, for recording the origin and edits of digital content. C2PA compares a credential to a nutrition label: a record of the content's history that anyone can inspect.

Does a missing label mean an image is real?

No. OpenAI's provenance guide says editing, converting or sharing a file can remove its C2PA metadata, while a SynthID watermark sits in the pixels or audio and "may survive some transformations". Meta says there are ways to strip invisible markers, and that it is not yet possible to identify all AI-generated content. Treat a found credential as evidence about origin, and a missing one as no evidence either way. For what the eye can and cannot catch, see the signature.

Sources

Checked 13 November 2026: