Pangram now spots AI images and videos as you browse

Short answer

Pangram, a New York startup best known for detecting AI-written text, launched Pangram Image 1.0 on November 4, 2026. It checks whether a picture was made or edited with AI, can show where in the image the AI part sits, and now runs inside the company's Chrome extension so it can flag AI images while you browse. A faster sibling model, Pangram Image Flash 1.0, checks videos frame by frame. All accuracy numbers below come from Pangram's own tests.

What happened

On its official blog, Pangram said Image 1.0 is out of research preview and available on the web, through its API and in the browser extension. The company says the model:

• catches 99.2% of AI images while wrongly flagging 0.27% of real ones (Pangram, via Engadget and The Next Web);
• scored 99.96% on a Pangram-built test of 1,200 AI "mirrors" of 100 real photos, made with 12 leading image generators;
• draws a heatmap, so a real photo with one AI-added object can be told apart from a fully generated picture;
• now accepts images as small as 256 × 256 px (down from 512), with a reduced-confidence note below 512 px.

For video, Pangram Image Flash 1.0 scored 98.7% on DF26, an academic benchmark of clips of people talking to camera, which Pangram describes as a 29-point lead over the next detector it tested. You can upload a video on Pangram's site and see which parts come back as AI.

Availability: Engadget reports the full image and video tools are for Pangram subscribers, while a free account can check a handful of images. The test sets were designed by Pangram, not by outside reviewers (The Next Web).

Why people are talking about it

AI pictures and videos now look good enough that most of us cannot tell at a glance. Pangram's own blog cites the DF26 study, which found people were barely better than chance at spotting AI talking-head videos. A tool that quietly labels images in your feed, on shopping pages or on rental listings is easy to understand and easy to try, which is why the launch was picked up by Engadget, Digital Trends and The Next Web on the same day.

What it means if you use AI

• AI images are not the problem; knowing what you're looking at is. Plenty of AI art is creative and fun. A clear label simply lets you enjoy it for what it is.
• Phones blur the line too. Pangram CEO Max Spero noted that most phone cameras, iPhones included, already use some AI processing on real photos, so the model had to learn the difference between AI touch-ups and fully generated images (Engadget).
• Treat it as a second opinion. No detector is perfect, and small or heavily compressed images are harder to judge.

Quick checklist

1. Try a free check on Pangram's image detector page with a picture you already know is real or AI.
2. If you browse in Chrome, look at the extension's labels on social feeds or listings.
3. Use the heatmap to see whether only part of an image was edited.
4. For low-resolution images, take the "reduced confidence" note seriously.
5. When something matters (a purchase, a rental), combine the result with common sense and a reverse image search.

Sources

• Pangram (Nov 4, 2026): Introducing Pangram Image 1.0 (primary)
• Engadget (Nov 4, 2026): Pangram expands its AI detection tools to video and images
• Digital Trends (Nov 4, 2026): Pangram releases Image 1.0 tool to sniff fake pics and videos
• The Next Web (Nov 4, 2026): Pangram's new models spot AI fakes in images and video, it says

Sources read on November 5, 2026 (Europe/Prague).

Related reading:

• ChatGPT Images 2.6: GIFs and 50% faster images
• Is ChatGPT safe? What happens to what you type

Cover photo: Avi Richards, "Editing with a View", via Wikimedia Commons (Unsplash), [CC0](https://creativecommons.org/publicdomain/zero/1.0/).