Detect AI
Content
100% Automatically and Free
100% Automatically and Free
EXIF presence, camera/codec traces, JPEG Q-tables, and missing/edited fields.
Keyword hits and phrasing patterns often associated with model output.
Edge density, Laplacian variance, blockiness—flat or overly smooth regions can be suspicious.
Error level analysis and low-ELA regions that may indicate synthetic composition.
Distribution and uniqueness checks (e.g., chroma/luma ratios, brightness/saturation moments).
Repetition, token variety, and sentence length distribution for text-based content.
The confidence score reflects the strength of observed signals, not a definitive ground truth. Some benign workflows (like exports or aggressive compression) can look AI-like, while sophisticated generations may evade detection. Always combine results with human judgment and source context.
Your last 15 uploads are stored only in your browser for convenience (see Uploads). For production deployments, set storage/retention policies to match your organization’s requirements.