Pangram Labs Launches Image Detection Model

US startup Pangram Labs announced Pangram Image 1.0 on November 4 local time, saying the AI model detects 99.2% of AI-generated images and has a 0.27% false-positive rate. Its Google Chrome extension can identify whether images are AI-generated while users browse social media and e-commerce sites.

In July, the company released the research preview Pangram Image 0.1. Version 1.0 incorporates user feedback and, according to Pangram Labs, improves detection of deepfakes, low-resolution images and images generated by state-of-the-art models. In a benchmark using 1,200 synthetic images made from real photos, the model detected 99.96%; the company also reported detection rates above 99% in several other tests.

Pangram Labs also released Pangram Image Flash 1.0, a fast model for detecting AI-generated videos. The company says it took first place on the DF26 video detection benchmark with a score of 98.69, 29 points ahead of the runner-up. The detector marks suspected AI-generated sections in red on a heat map.

The Pangram AI Detection extension is free to install on Chrome, though some features, such as automatic social media scanning, require a paid plan. The models are also available through an API. Pangram Image 1.0 costs $0.25 per image; Pangram Image Flash 1.0 costs $0.05 per image and $0.05 per second of video. The detection rates and benchmark results are based on the company’s statements, and the article does not provide detailed test conditions. It also notes that the extension’s basic use and automatic scanning have different pricing, and that installing the extension for free does not grant access to paid features. The article provides no further pricing details. Detection on social media and e-commerce sites is done through the browser extension. Although the video model’s name includes “Image,” it is offered alongside the image model. Users can access both models through the extension or API.

For the video model, Pangram Labs cited its DF26 score and lead over the runner-up; for the image model, it reported results from several benchmarks. The figures reported are based on the company’s account.

US startup Pangram Labs announced Pangram Image 1.0 on November 4 local time, saying the AI model detects 99.2% of AI-generated images and has a 0.27% false-positive rate. Its Google Chrome extension can identify whether images are AI-generated while users browse social media and e-commerce sites.

In July, the company released the research preview Pangram Image 0.1. Version 1.0 incorporates user feedback and, according to Pangram Labs, improves detection of deepfakes, low-resolution images and images generated by state-of-the-art models. In a benchmark using 1,200 synthetic images made from real photos, the model detected 99.96%; the company also reported detection rates above 99% in several other tests.

Pangram Labs also released Pangram Image Flash 1.0, a fast model for detecting AI-generated videos. The company says it took first place on the DF26 video detection benchmark with a score of 98.69, 29 points ahead of the runner-up. The detector marks suspected AI-generated sections in red on a heat map.

The Pangram AI Detection extension is free to install on Chrome, though some features, such as automatic social media scanning, require a paid plan. The models are also available through an API. Pangram Image 1.0 costs $0.25 per image; Pangram Image Flash 1.0 costs $0.05 per image and $0.05 per second of video. The detection rates and benchmark results are based on the company’s statements, and the article does not provide detailed test conditions. It also notes that the extension’s basic use and automatic scanning have different pricing, and that installing the extension for free does not grant access to paid features. The article provides no further pricing details. Detection on social media and e-commerce sites is done through the browser extension. Although the video model’s name includes “Image,” it is offered alongside the image model. Users can access both models through the extension or API.

For the video model, Pangram Labs cited its DF26 score and lead over the runner-up; for the image model, it reported results from several benchmarks. The figures reported are based on the company’s account.