How OpenAI’s DALL‑E 3 Watermarks Reveal AI‑Generated Images
OpenAI’s DALL‑E 3 will embed both invisible metadata and a visible “CR” watermark in generated images, allowing users to verify AI origins via C2PA credentials, with minimal impact on speed or quality, though the marks can be removed and are currently limited to static images.
OpenAI announced that its image generator DALL‑E 3 will start adding watermarks from the Content Credentials and C2PA to generated images, helping users identify AI‑generated content.
The watermark will appear in images generated on the ChatGPT website and DALL‑E 3 model API; mobile users will see the watermark starting February 12.
The watermark consists of two parts: an invisible metadata component and a visible “CR” symbol placed in the upper left corner of each image.
Users can verify the source of any image generated by the OpenAI platform via sites such as Content Credentials Verify. Currently only static images support the watermark; video and text content do not.
OpenAI says adding watermark metadata has negligible impact on generation speed and image quality, though it may slightly increase file size for some tasks.
C2PA, formed by Adobe, Microsoft and others, aims to identify content provenance through content‑credential watermarks and indicate whether content was created by a human or AI.
However, the watermark is not a foolproof method to stop misinformation. OpenAI notes that C2PA metadata can be easily removed accidentally or deliberately, especially since most social media platforms strip metadata from uploads, and screenshots lack metadata.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
21CTO
21CTO (21CTO.com) offers developers community, training, and services, making it your go‑to learning and service platform.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
