New AI Tool Detects Deepfakes with 91% Accuracy

A powerful new weapon just arrived in the fight against deepfake images that have devastated lives and enabled billions in fraud.

Researchers at Germany's Fraunhofer Institute have developed RealorRender, a detection system that identifies fake images with accuracy rates between 85% and 91%. Unlike previous tools, this one doesn't just flag suspicious images—it shows exactly why it thinks they're fake.

The timing couldn't be more critical. In 2023, investigators found over 276,000 sexual deepfakes on a single website, with 96% targeting women. That same year, so-called "nudification" sites attracted 24 million visits in September alone, and deepfake-related fraud topped $1.5 billion in 2025.

RealorRender works by first trying to recreate a suspicious image using AI generation tools. If the AI successfully clones the image, that very success proves the original was likely a deepfake. Then the system highlights problem areas using colored heatmaps, showing users the specific textures and patterns that reveal synthetic origins.

"We use explainable AI methods for this, which allows us to highlight the image areas and structures that contribute to the model's decision," says Nadia Burkart, who leads the Applied Explainable AI group at Fraunhofer. This transparency helps developers continuously improve the system's accuracy.

Senior scientist Andreas Specker explains the hybrid approach: "First, we use an AI image generator to reconstruct the image. Then an AI model takes over the classification and carries out a hybrid calculation of the reconstruction error." The result is a percentage showing how confident the system is in its assessment.

The technology arrives as governments worldwide struggle to address the deepfake crisis. UK Minister Jess Phillips recently pledged action, saying "Every child deserves to grow up safe, and we will do whatever it takes to make that a reality." But policy alone can't solve the problem without detection tools that actually work.

The Ripple Effect

RealorRender's impact extends far beyond individual safety. Banks and financial institutions can use it to prevent fraud. Schools can protect students from harassment and exploitation. Social media platforms can block harmful content before it spreads. Political campaigns can verify authentic footage during elections.

The system's transparency also builds trust in ways previous detection tools couldn't. When RealorRender flags an image, it doesn't just say "this looks fake"—it shows users the specific elements that triggered the alert. That accountability makes the technology more reliable and helps train both humans and other AI systems to spot fakes.

The German Federal Office for Information Security funded the project, recognizing deepfake detection as critical infrastructure for the digital age. As deepfake technology becomes more sophisticated, having detection systems that can explain their reasoning becomes increasingly vital.

For the millions already targeted by deepfakes and the countless others at risk, this breakthrough means something simple but profound: a real chance at digital safety.