The AI labPangram Labs, which focuses on AI-generated content recognition, officially released the Pangram Image1.0 image detection model and the Pangram Image Flash1.0 video detection model on November 4. The debut of this series of tools marks a higher-precision technological breakthrough in digital content authenticity verification.

In terms of core performance, Pangram Image1.0 used 12 cutting-edge AI image generators and built a test dataset of 1,200 photos based on 100 real images. Its test accuracy reached 99.96%, and it uses heatmaps to precisely locate AI-generated content.

At the same time, the model has excellent robustness and can effectively resist interference from compression, scaling, and manual edits to its detection capability. It still maintains 96.4% accuracy even at a low resolution of 256×256. In addition, the Pangram Image Flash1.0 video detection model unveiled the same day achieved 98.7% accuracy on the DF26 dataset, leading the runner-up by 29 percentage points.

With the rapid development of generative AI, efficiently identifying fake visual content has become a core challenge in digital content security governance. The release of Pangram's new-generation detection models not only raises the industry's performance ceiling for image and video authenticity detection, but also provides more reliable technical support for the entire industry to address the misuse of AI technology and maintain the authenticity of the digital ecosystem.