AI DEEPFAKE DETECTOR
VERIFY IF AN IMAGE IS REAL
IS THIS REAL?
Scan any image or video for AI manipulation.Upload an image or video to detect AI manipulation.
or paste from clipboard (⌘V / Ctrl+V)
Detects content from
How Our Deepfake Detection Engine Works
Our multi-modal forensic pipeline analyzes invisible mathematical artifacts embedded in every image and video to determine whether the media was captured by a real camera or generated by an AI model such as Midjourney, DALL-E 3, Stable Diffusion, Flux, or Sora. We do not rely on visual inspection alone - we analyze the physics of the image itself.
GAN Fingerprinting
Every AI generator - Midjourney, Flux, DALL-E 3, Stable Diffusion - leaves a unique mathematical signature in the frequency domain of its output images. Our engine identifies these specific generator artifacts to attribute synthetic content back to its source model, even after resizing or light re-compression.
Error Level Analysis (ELA)
We analyze JPEG compression artifacts at the pixel block level. Real photographs have a uniform compression history across the entire image. AI-generated regions, composited faces, and inpainted areas compress differently - ELA maps those inconsistencies into a visual anomaly report that makes manipulation visible.
Metadata & EXIF Audit
Every camera-captured photo embeds rich EXIF metadata: GPS coordinates, shutter speed, ISO, lens model, and a timestamp. AI-generated images are born without this hardware record. We flag stripped, incomplete, or inconsistent EXIF data as a strong forensic signal of synthetic origin.
Frequency Domain Analysis
A real camera lens acts as a natural low-pass filter, producing a smooth frequency rolloff. AI diffusion models generate high-frequency energy spikes in the Fourier spectrum - unnatural clusters that result from the upscaling and de-noising steps of generative inference. We detect these spectral anomalies automatically.
Multi-Signal Evidence Fusion
No single forensic signal is definitive on its own. A re-compressed image may weaken ELA evidence. A well-crafted synthetic image may mimic noise patterns. That is why FauxLens operates a multi-stage evidence fusion pipeline - combining GAN fingerprinting, ELA, noise floor analysis, EXIF inspection, shadow logic, and frequency anomaly detection into a single weighted confidence score. The result is a probability estimate supported by multiple independent lines of forensic evidence, making it far harder to defeat than any single-signal detector.
Why Choose FauxLens?
Privacy-First Architecture
Zero-retention policy. Your images are processed in RAM and permanently discarded after analysis. We never store, share, or train on your uploads.
Covers All Major Models
Detection signatures for Midjourney v6, DALL-E 3, Stable Diffusion XL, Flux, Sora, and Veo 3 - updated continuously as new generators emerge.
Evidence, Not Verdicts
We report confidence scores and specific anomalies - not binary labels. Use FauxLens as a forensic instrument in a broader verification workflow.