New tool · November 5, 2026 · 2 min read · Source: Pangram Labs

Pangram Image 1.0: new tool detects AI images and videos

On November 4, 2026, Pangram Labs launched Pangram Image 1.0 and Image Flash 1.0. The two models are designed to detect AI-generated images and videos and to show which parts of an image were made by AI.

My tool tip · AdWispr Flow – AI dictationTry it free

Fictional woman wearing clear glasses with a glowing pixel grid on the lenses, green and blue light (AI image) — Illustration

Key facts

Pangram Labs officially launched Pangram Image 1.0 (images) and Image Flash 1.0 (video) on November 4, 2026.
The vendor reports 99.96% accuracy for Image 1.0 on its own test set and 98.7% for Flash 1.0 on DF26; neither has been independently verified.
Available via the website, a Chrome extension and an API; images from 256 × 256 pixels can be checked.

What's new

Pangram is best known for detecting AI-written text. Its image detector has been in preview for a few months and is now officially out:

Pangram Image 1.0 checks images and can show a heatmap marking AI-edited areas, so it also covers photos where only part was changed.
Image Flash 1.0 is a smaller, faster model for video. It analyzes a clip frame by frame and flags the sections that contain AI material.
Smaller images: the minimum size drops from 512 × 512 to 256 × 256 pixels. Below 512 pixels, the tool itself warns that confidence is lower.

Where you can use it

You can upload images and videos on Pangram's website. There is also a Chrome extension meant to flag AI content while you browse social media, shopping or rental sites, plus an API for businesses. The announcement gives no prices and points to the general pricing page.

How well it works

Pangram says Image 1.0 scored 99.96% on its own test set of 100 real photos and 1,200 AI versions from 12 image generators. The video model reached 98.7% on the DF26 benchmark, which the company says is 29 points ahead of the next-best detector. heise notes that these numbers have not been independently verified. The DF26 study also shows how hard the task is: people spot AI videos of talking people barely better than chance.

What this means for you

Anyone who shares images or videos gets another way to check them, but not proof.
The heatmap helps with partly edited photos that are hard to spot by eye.
The high accuracy figures are vendor claims; false alarms remain possible.

How to use this

If a suspicious image or video matters for a decision, run it through a detector and ask for the original source at the same time. Never rely on a detector's result alone.

Luigi's take

AI images are now good enough that you can't trust your own eyes. I use AI images on this website and label them, and that kind of honesty matters more as detection tools spread. Anyone passing off AI images as real photos is increasingly likely to get caught.

My advice: treat these tools as a second opinion, not a verdict. When an image matters for a decision, like a receipt, a rental listing or a supposed product photo, also ask for the source. And for your own content, label AI images openly and you'll have nothing to fear from a detector.