Technology

A New Tool Checks Whether Pictures and Videos Were Made by AI

Martin HollowayPublished 2w ago3 min readBased on 9 sources
Reading level
A New Tool Checks Whether Pictures and Videos Were Made by AI
source:pangram.com

Pangram, a company known mainly for detecting AI-written text, has released tools that flag AI-generated images and video. Paying subscribers can upload a video and get an assessment of whether AI was used to create or edit it, according to Engadget, which reported the launch on November 4, 2026.

The video tool sorts clips into two groups: those that are partly AI-generated and those that are entirely AI-generated. It also includes a heatmap, a colored overlay that shows which parts of a video are likely made by AI. Pangram says the model is especially good at spotting the newest AI video, and that its own testing found more than 98 percent accuracy in flagging AI content.

Image support comes as Pangram Image 1.0, which works through the company's Chrome extension. The extension can scan pictures automatically as you browse and flag the ones that are likely AI-generated. Pangram says its image model caught more than 99 percent of AI images in testing. It wrongly flagged real images in 0.27 percent of cases.

Results for images smaller than 512 x 512 pixels come with a warning that the tool is less confident. This matters in everyday use, because many pictures in social media feeds and thumbnails are that small.

Access is tiered. The new image and video features are available now to subscribers. A free account can check a handful of images without paying. The free text checker, which covers output from ChatGPT, Claude and Gemini, has not changed.

Pangram's API has also been updated, along with the release of Flash 1.0, to support image and video detection. The company says its image detection covers outputs from major providers, including OpenAI's GPT Image. Its video detection covers Kling, Seedance, Veo and Wan. The details are in its blog post on Pangram Image 1.0 and its image detection announcement.

Pangram was founded in Brooklyn, New York, in 2023. In late July, it announced $9 million in funding from Menlo Ventures to expand from text to images, as described in a Business Wire release. The company says it is trained to detect the latest models from OpenAI, Anthropic, Google, Meta, DeepSeek and xAI. It also claims that independent researchers ranked it the most accurate AI detector. Researchers at the University of Chicago and the University of Maryland have verified its accuracy, the company says. Its website says independent studies found it beats trained human readers at spotting AI-generated content.

In my view, the limits deserve as much attention as the headline numbers. A detector learns from examples made by particular AI systems, so its accuracy describes those systems. Think of it as a guard who learned to spot forgeries by studying last year's examples. The real test is how it handles pictures and videos from AI systems released after its training ended. Pangram says it will keep retraining the detector on new models, which is the only realistic way to keep up with tools that change every few months.

The broader context here is a shift in where checking happens. Detecting AI text has been a concern in schools and publishing for years. Image and video checks move that question into the web browser, where an ordinary person can test a suspicious picture before sharing it. That is useful for anyone who cannot tell a fake photo from a real one by looking. It also means the tool's errors, including its weaker results on small or compressed files, now affect everyday decisions by people who have no interest in how the model works.