Pangram, which builds AI-content detectors, announced Pangram Image 1.0 and Pangram Image Flash 1.0 on November 4. Image 1.0 is its flagship detector for AI-generated still images. Image Flash 1.0 is a smaller, faster model that checks videos frame by frame.
Pangram says Image 1.0 scored 99.96% on an evaluation set of 1,200 synthetic images, made by producing AI "mirrors" of 100 real photos with 12 frontier image generators and prompts that closely matched each original. The detector produces heatmaps showing which parts of an image contain AI content, and the company claims it holds up under edits, compression and low resolution. The minimum input size drops from 512x512 to 256x256 pixels, with a reduced-confidence notice for images under 512 pixels.
For video, Pangram reports that Image Flash 1.0 scored 98.7% on the DF26 benchmark, which it calls a 29-point lead over the next-best detector. Its industry comparison used 1,529 videos, including 129 real social media clips and AI mirrors from 14 video generators. All figures come from Pangram's own evaluations. The post does not report a false-positive rate and does not publish competitors' scores in its text.
Image detection is available at pangram.com/image-detection and video upload at pangram.com/video-detection. Image 1.0 is also built into Pangram's Chrome extension, and the Pangram API now accepts both images and video. The announcement does not mention pricing.
For platforms that moderate user uploads, API access to image and video detection in one service makes automated screening easier to set up. Without an independent false-positive figure, though, teams should test it on their own content before using the scores to take action.