Let you identify AI-generated images! OpenAI uses SynthID watermarks and has launched a public verification tool.

OpenAI announces enhanced AI image recognition, introducing not only the C2PA certificate but also the first-time integration of Google's developed SynthID invisible watermark, along with a public verification tool to accelerate the establishment of a cross-company common provenance standard.

On May 19, OpenAI announced that it aims to make it easier for the public to identify "images generated by AI." In addition to the existing C2PA content certificates, OpenAI will now embed an additional layer of SynthID invisible watermark into images and launch a public verification tool that anyone can upload images to check if they originate from OpenAI's models. SynthID is not an OpenAI proprietary technology but a watermark standard developed by Google DeepMind.

Why use both C2PA and SynthID mechanisms simultaneously

C2PA content certificates involve embedding metadata within image files to record their source and editing history; however, this metadata is often removed when uploading to social platforms, saving, or taking screenshots. SynthID, on the other hand, embeds a set of signals invisible to the human eye directly into the pixels, allowing detection even if the metadata is stripped. Running both mechanisms in parallel ensures that when metadata is preserved, the full provenance can be verified, and even if it is removed, a recognizable watermark remains.

Verification tool launched, but currently only recognizes images from OpenAI

Along with this mechanism, a public verification tool has been released, allowing anyone to upload images to confirm whether they were generated by OpenAI's models. The watermark is initially applied to images produced via ChatGPT, Codex, and the OpenAI API. However, the current recognition scope of the tool is limited to OpenAI's own products, unable to identify images generated by other companies' AI tools. OpenAI states that they hope to gradually expand the coverage in the future.

Google and OpenAI align on SynthID

SynthID was developed by Google DeepMind, originally mainly used to mark content generated by Google’s own AI. Now, OpenAI has adopted the same watermark standard, followed by Kakao and speech company ElevenLabs, with Nvidia having done so earlier. As the two most prominent AI companies converge on the same provenance marking system, the approach to labeling and verifying AI-generated content is gradually becoming a cross-company standard.

• This article is reprinted with permission from: 《Chain News》
• Original title: 《OpenAI Adds Provenance to AI Images: Adopts Google SynthID Watermark and Launches Verification Tool》
• Original author: Elponcrab