Pangram launches AI detectors for images and videos

Expanding beyond text

Pangram, a startup known for its text-based AI detection, has officially launched tools to identify AI-generated images and videos. This move comes as generative AI becomes more common across various forms of digital communication. The company aims to provide a way for users to separate authentic content from what they call 'AI slop.'

The new tools allow users to upload video files to receive an assessment of whether AI was used to create or edit the clips. While the video and upgraded image tools are primarily for paid subscribers, free users can still test a few images. This expansion follows Pangram's success in the text space, where it has previously outperformed rivals in accuracy.

According to a recent analysis by the digital marketing firm Graphite, roughly 50 per cent of news articles and blogs on the internet are now AI-generated. By moving into visual media, Pangram is positioning itself to address a significant portion of the web's synthetic content. The company is joining a growing group of startups trying to solve the problem of distinguishing human work from machine output.

Technical hurdles in detection

Building a reliable model for images and videos is significantly more complex than detecting AI-generated text. Pangram CEO Max Spero noted that there are many more AI image and video models available than there are frontier large language models. Because these models produce such diverse outputs, creating a consistent detection standard is difficult.

One specific challenge the company faced involved the way modern smartphones process photos. Most phone cameras, including those from Apple, use AI post-processing to add detail to real-life images. Pangram had to ensure its model was robust enough to recognize these standard edits without incorrectly flagging real photos as fully synthetic.

To solve this, Spero explained that the company delayed the release of its image detection model to test it on several iPhone 18 Pros. The initial version of the model had mistakenly flagged images from these cameras as AI-generated. By conducting this testing, the company worked to understand the nuanced difference between simple AI edits and visuals that are entirely created by AI.

How the new tools work

The upgraded image detection model, known as Pangram Image 1.0, is now available through a dedicated Chrome extension. This tool can automatically scan images as people browse the internet and flag those that are likely to be AI-generated or AI-modified. The company reports that the model flagged more than 99 per cent of AI-generated images in its testing.

For video content, the new tool is designed to distinguish between clips that are 'partially AI' and those that are entirely AI-generated. It provides users with a heatmap that highlights the specific parts of a video that appear to be modified by artificial intelligence. This visual aid helps users see exactly where the synthetic elements are located within a clip.

Pangram states that the underlying model for video detection particularly excels at identifying content from frontier AI video models. In internal testing, the tool was able to flag AI-modified content with more than 98 per cent accuracy. These high accuracy rates are intended to provide a reliable baseline for users trying to verify the authenticity of the media they consume daily.

Accuracy and limitations

While the company reports high success rates, there are specific technical limitations to the current image detection tool. For instance, the results will come with a reduced confidence disclaimer when the images have resolutions below 512 x 512. This suggests that lower-quality images may be harder for the model to analyze with certainty.

The company also maintains a very low false positive rate of 0.27 percent for its image detection testing. This means that the tool is designed to be highly conservative, aiming to avoid labeling real human-made content as AI. This balance is crucial for maintaining trust with users who rely on the tool for verification.

Pangram's reputation for accuracy was established in the text space before these visual tools were released. A 2025 study by the University of Chicago found that Pangram beat rivals like Originality.ai and GPTZero in assessing texts of roughly 500 words or more. The company is now attempting to bring that same level of near-perfect accuracy to the much more complex world of images and video.

The future of verification

The launch of these tools highlights a growing need for verification as generative AI continues to permeate communication. As more news and media are produced by machines, the ability to distinguish between reality and synthesis becomes a vital skill. Pangram's expansion suggests that visual detection will become a standard feature for many web browsers and content platforms.

For the average user, these tools offer a layer of protection against misinformation and 'AI slop.' By providing heatmaps and automated scanning, Pangram is moving toward a future where AI detection is integrated into the browsing experience. This allows people to make more informed decisions about the content they see on their screens.

Moving forward, the competition among startups to create the most accurate detectors will likely intensify. As AI models become more sophisticated, detection tools will need to evolve even faster to keep up. Pangram's focus on testing against real-world hardware like iPhones shows that the next phase of detection will involve navigating the blurry line between human photography and machine generation.