Is This Image AI-Generated?

About this detector

The optional local classifier is Community Forensics DeepfakeDet-ViT (ViT-Small/16, 384px), MIT-licensed, trained on about 2.7 million images from thousands of generators (Park and Owens, CVPR 2025). This site loads the official ONNX FP32 export in your browser with ONNX Runtime Web (WebGPU where available, otherwise WASM). First use downloads the weights (about 83 MB) to your browser cache. Quantized INT8 weights are not used because ONNX Runtime Web cannot run their ConvInteger operators. There is no universal detector for every modern generator.

DevNestro does not detect Google SynthID or other proprietary watermarks. No official SynthID SDK is integrated.

How to use it

1. Drop a JPG, PNG, or WebP (max 20 MB).
2. Review the preview, then choose Analyze Image. Heavy model download starts only then.
3. Read the overall category, then the four evidence cards. Expand technical details if you need model and file facts.

FAQ

Can DevNestro tell with 100% certainty whether an image is AI-generated?

No. No universal detector can reliably determine the origin of every image. DevNestro combines provenance, metadata and machine-learning signals to provide an evidence-based assessment.

What are Content Credentials?

Content Credentials are provenance records, often stored as a C2PA manifest, that can name a tool, an editing history, or an AI disclosure when the creator attached them. This version reports presence and readable labels; it does not verify cryptographic signatures.

Does no Content Credential mean the image is real?

No. Credentials can be unavailable, removed, or unsupported.

Is my image uploaded?

No. Normal analysis runs in the browser. Model files may be downloaded to the browser from the model host on first use.

Does DevNestro detect Google SynthID?

No. DevNestro does not claim to detect proprietary watermarks unless an official supported detector is integrated.