Pangram has transitioned its AI detection capabilities from a research preview to a production-ready tool with the release of Pangram Image 1.0. The platform is designed to identify synthetic media across various formats, offering a flagship model for images and a specialized, faster model called Pangram Image Flash 1.0 for video content. By providing these tools, Pangram aims to help users distinguish between real photographs and AI-generated content in environments like social media feeds and online marketplaces.
The core mechanism of Pangram Image 1.0 involves identifying specific artifacts within images to determine their origin. A key feature of this release is the inclusion of a heatmap output, which allows the model to pinpoint the exact areas within a file where AI-generated elements appear. To ensure reliability, the model was trained to be robust against common digital transformations, such as compression, blurring, and noise—edits frequently used by social media platforms or by actors attempting to bypass detection systems. During internal testing against 1,200 synthetic images created from 12 frontier AI generators, the model achieved a reported accuracy of 99.96%.

The image demonstrates the core mechanism of the tool, specifically the heatmap feature used to localize synthetic artifacts.
For video content, the company introduced Pangram Image Flash 1.0, which performs frame-by-frame analysis to detect AI-edited segments. According to Pangram, this model scored 98.7% on the DF26 academic benchmark, which they claim is a 29-point improvement over the next best detector. While the system offers high performance, there are specific technical limitations: while the minimum input size was previously 512x512, the new model supports resolutions as low as 256x256. However, for any images processed below the 512x512 threshold, the system provides a reduced confidence disclaimer to account for the potential drop in detection accuracy at lower resolutions.

This visual supports the claim regarding the model's high accuracy and performance relative to other detectors.