OpenAI is working on new AI image detection tool

OpenAI has added a new tool to detect whether an image was created with its DALL-E AI image generator, as well as new watermarking methods to more clearly mark the content it generates Are.

In a blog post, OpenAI announced that it has begun developing new provenance methods to track content and prove whether it is AI-generated. These include a new image detection classifier that uses AI to determine whether the photo was AI-generated, as well as a tamper-resistant watermark that can tag content like audio with invisible signals.

The classifier predicts the probability that an image was created by DALL-E 3. OpenAI claims that the classifier works even when the image is cropped or compressed or the saturation is changed. While the tool can detect whether images were created with DALL-E 3 with about 98 percent accuracy, its performance in detecting whether the content was from other AI models is not as good as other image generators. . Only 5 to 10 percent of the paintings have been identified. , like Midjourney.

OpenAI previously added content credentials from the Coalition of Content Provenance and Authority (C2PA) to image metadata. Content credentials are essentially watermarks that include information about who owns the image and how it was created. OpenAI, along with companies like Microsoft and Adobe, is a member of C2PA. This month, OpenAI also joined the C2PA steering committee.

The AI company has also started adding watermarks to clips from the voice engine, its text-to-speech platform currently in limited preview.

Both the image classifier and the audio watermarking signal are still being refined. OpenAI says it needs to get feedback from users to test its effectiveness. Researchers and nonprofit journalism groups can test the image detection classifier by implementing it on OpenAI's Research Access platform.

OpenAI has been working on AI-generated content detection for years. However, in 2023, it had to end a program that attempted to identify AI-written text because the accuracy of the AI text classifier was consistently low.