top of page
  • Writer's pictureSarah Ruivivar

OpenAI Develops Tools to Detect AI-Generated Images

Image credits: The Verge

OpenAI, the leading artificial intelligence company, is stepping up its game with new tools designed to detect if an image has been generated by its DALL-E AI image generator.


In addition, the company is introducing innovative watermarking methods to flag AI-generated content more clearly.


In an exciting blog post, OpenAI revealed it's in the process of developing advanced provenance methods to track content and verify if it was AI-generated. This includes a cutting-edge image detection classifier that utilises AI to determine whether a photo is AI-generated. Additionally, a tamper-resistant watermark is being developed that can invisibly tag content such as audio.


The image classifier is designed to predict the likelihood of a picture being created by DALL-E 3. OpenAI boasts that the classifier maintains its functionality even if the image is cropped, compressed or the saturation is altered. While the tool can detect DALL-E 3 images with approximately 98% accuracy, its performance with content from other AI models is less impressive, identifying only 5 to 10% of images from other generators like Midjourney.


 

Want to learn more about AI's impact on the world in general and property in particular? Join us on our next Webinar! Click here to register

 

OpenAI has previously incorporated content credentials into image metadata from the Coalition of Content Provenance and Authority (C2PA). These credentials are essentially watermarks containing information about the image's ownership and creation process. OpenAI, along with tech giants such as Microsoft and Adobe, is a member of C2PA and recently joined its steering committee.


Moreover, OpenAI has started adding watermarks to clips from Voice Engine, its text-to-speech platform currently in limited preview.


Both the image classifier and audio watermarking signal are still under refinement. OpenAI emphasises the need for user feedback to test their effectiveness. Researchers and non-profit journalism groups are invited to test the image detection classifier via OpenAI’s research access platform.


While OpenAI has been striving to detect AI-generated content for years, it had to terminate a program in 2023 that aimed to identify AI-written text due to consistent low accuracy. With these new developments, OpenAI is once again at the forefront of ensuring transparency and authenticity in the world of AI-generated content.



 

Want to learn more about AI's impact on the world in general and property in particular? Join us on our next Webinar! Click here to register

 


Made with TRUST_AI - see the Charter: https://www.modelprop.co.uk/trust-ai

23 views0 comments

Recent Posts

See All

Comments


bottom of page