Free Online AI Detector for Text and Images

1. 1 · Watermarks

   SynthID and Content Credentials

   Google and OpenAI stamp an invisible SynthID watermark into the pixels of their images, and many tools, OpenAI included, add a C2PA Content Credential to the file. A credential names the generator, and a signed one from OpenAI, Microsoft, or Adobe is checked against a bundled trust list, so a forgery fails instead of passing as proof. A SynthID read names the provider instead, and it survives the screenshot that strips a credential.

2. 2 · Image check

   Pixel-level image model

   Take a screenshot and the C2PA credential may disappear. That's the problem with file metadata: it can be cut away from the image. The pixels remain, so we check those next. Slop or Not's AI image detection model looks for traces left by image generators. In our internal tests, it was about 90% accurate on non-watermarked images.

3. 3 · Text model

   Transformer trained on human and AI writing

   A fine-tuned transformer trained on a large corpus of human and AI writing scores the passage for AI fingerprints. On the public RAID benchmark, it caught 99.5% of AI text at a 5% false-positive rate, adversarial attacks included. Full, natural sentences give a more confident result than fragments or single lines.

Image accuracy figures come from our internal tests; text detection is independently scored on RAID. Results can vary with new AI models and methods built to trick detectors.

See the full RAID benchmark results