Pangram releases Image 1.0 to detect AI-generated photos and videos

High-accuracy detection tools

Pangram, a company known for its AI text detection, is expanding its technology to include multimedia analysis. The company is launching Pangram Image 1.0, a tool designed to identify pictures created by artificial intelligence. This tool was previously only available as a research preview but is now open to the public.

The company claims that its image detector can identify AI-generated pictures with 99.96% accuracy on its own evaluation set. It has been specifically updated to handle difficult cases, including low-resolution images and those edited by humans. It also performs well against pictures produced by the most advanced AI models currently available.

Users can access these tools through the company's website, an API, or a new Chrome extension. The extension is designed to work while users browse the internet, automatically flagging AI content on social media platforms, online shopping sites, and rental listings. As digitaltrends.com reported, the company aims to provide a standard for verifying digital media.

How the technology works

The Image 1.0 model is trained to recognize subtle artifacts that often appear in synthetic media. This includes identifying patterns caused by compression, blurring, and artificial noise that typically fool other AI detectors. By recognizing these technical flaws, the model can distinguish between human-made and machine-made content.

One specific feature of the tool is its ability to generate a heatmap. This allows users to see exactly which parts of a specific image were flagged as AI-generated rather than just giving a general score. The system has also been updated to handle smaller files, now checking images as small as 256 x 256 pixels.

For moving content, the company uses a separate model called Pangram Image 1.0 Flash. This is a lighter-weight version designed to perform frame-by-frame detection on videos. This method allows the system to maintain high accuracy, reaching 98.57% on AI-generated videos, which is significantly higher than several other commercial competitors.

Outperforming the competition

Pangram compared its new tools against several other commercial detectors and academic research models. In these tests, the company's tools consistently outperformed the benchmarks set by researchers. The company highlights its lead over other products like TruthScan, AI or Not, and Sightengine.

To verify the video detector's performance, the company conducted a specific test using 129 real videos from social media. They created AI-generated versions of these videos using 14 of the latest AI video models to see if the detector could catch them. The results showed a significant lead over the next-best detector on the DF26 benchmark.

The company notes that while it is highly accurate, images smaller than 512 pixels will trigger a reduced-confidence warning. This ensures that users are aware of the limitations of the tool when analyzing very low-quality files. The goal is to provide a reliable standard for identifying synthetic content across various platforms.

Combating AI-driven scams

The release of these tools comes at a time when scammers are using AI to create more convincing fraud. One disturbing example involves criminals using AI-generated pet photos to target people who have lost their animals. These scammers pretend to have found the missing pet and then demand money from the desperate owners.

In some cases, these criminals use aggressive tactics, including making threats against the animals if the owners do not pay. Because AI can create such realistic images, these scams are becoming harder for the average person to spot without help. The new Pangram tools are intended to help users verify these types of emotional and high-pressure scams.

Scammers are also using AI to create highly personalized phishing emails. Researchers have found that AI can scrape public social media posts to learn about a person's recent activities, such as places they visited or things they are looking forward to. By using this personal data, scammers can create emails that feel much more authentic than standard spam.

What comes next for detection

The availability of Pangram Image 1.0 marks a shift toward more accessible AI verification tools for everyday internet users. By integrating the technology into a Chrome extension, the company is making it easier for people to see warnings in real-time as they scroll through their feeds. This could help reduce the spread of misinformation on large social networks.

As AI models continue to improve, the