OpenAI has confirmed it added SynthID watermarking to its image generator - embedding an invisible signal developed by Google DeepMind directly into every ChatGPT image output.
SynthID works by encoding a digital provenance marker into the pixel data of an image at the time of creation. The marker is invisible to the human eye but detectable by compatible tools. Crucially, it's designed to survive the edits people typically use to obscure AI origin: cropping, resizing, color adjustments, brightness changes, and screenshots. The goal is to make it possible to identify AI-generated images even after someone has tried to clean them up.
The fact that OpenAI is using Google DeepMind's technology is notable on its own. These are competing companies, and SynthID has been Google's watermarking standard, already deployed across Imagen and other Google image tools. Its adoption by OpenAI suggests the industry may be converging on SynthID as a shared baseline for AI image provenance rather than each company building its own incompatible system. The other major approach - championed by Adobe and the C2PA coalition (Coalition for Content Provenance and Authenticity) - works differently, attaching cryptographically signed metadata to image files rather than encoding signals into the pixels themselves.
For anyone producing AI-generated images for commercial use - ads, social media, editorial work, stock submissions - this matters more than it might seem right now. Publishers, ad platforms, and stock image sites are actively building detection pipelines to flag AI-generated content. A standardized embedded signal makes that detection more reliable than inferring origin from visual artifacts or metadata that can be stripped.
OpenAI has not specified whether SynthID applies to all image outputs or only certain resolution and quality tiers.