OpenAI announced thatImage Generator DALL-E 3 will begin adding watermarks from the Content Source and Authenticity Alliance (C2PA) to the generated images to help users recognize content generated using artificial intelligence (AI).
It is understood that the watermark will appear on the ChatGPT website and in images generated by the DALL-E 3 model API, and that mobile users will see the watermark from February 12th. The watermark consists of two parts: an invisible metadata component and a visible CR symbol, which is located in the upper left corner of each image.
Users can query the origin of any image generated by the OpenAI platform (which image was used) through websites such as Content Credentials Verify. AI Tools). Currently, only still images support watermarking; video and text content are not yet supported.
OpenAI says that adding watermarked metadata to an image has a negligible impact on generation speed and image quality, but may slightly increase the image size for some tasks.
C2PA, which consists of companies such as Adobe and Microsoft, aims to identify the source of content through content credential watermarking and show whether the content was generated by a human or an AI.
However, watermarking is not a foolproof method of stopping the spread of disinformation, and OpenAI notes that C2PA's metadata can be "easily deleted accidentally or intentionally," especially since most social media platforms routinely delete metadata from uploaded content. Screenshots also do not carry metadata.