Model Rankings

Ranking generative AI models by their ability to deceive the human eye.

ANALYZING MODEL PERFORMANCE...

Methodology & Insights

The Trick Rate Metric

The primary metric used for ranking is the Trick Rate. This represents the cumulative percentage of games where players incorrectly identified the AI-generated image as the real photograph. A high Trick Rate indicates that a model's generative capabilities are reaching a point of parity with real-world textures, lighting, and anatomical correctness.

Our Testing Framework

Each AI model is subjected to rigorous real-world testing. We feed the same high-resolution source photographs through an image-to-text-to-image pipeline for every model. This ensures that we are measuring the model's ability to recreate a specific scene rather than its ability to generate generic "cool" images.

Global Anomaly Heatmaps

By analyzing where users click when they successfully identify a fake, we build heatmaps of model-specific weaknesses. Whether it's the classic "AI hand" issue, inconsistent reflections, or texture blurring, our data reveals the unique fingerprint of each generative version.

Human-in-the-Loop Perception

Unlike synthetic benchmarks (like FID or CLIP scores), the Realdle Rankings rely entirely on human biological perception. This provides the most accurate assessment of how AI-generated content impacts our actual digital literacy and ability to discern truth in the modern web era.

Statistics are updated in real-time as global players contribute new data. Currently tracking performance across thousands of daily sessions.