DETECTION STUDIO

Is this image AI-generated? Get a clear answer.

Inspect any PNG, JPEG, WebP or GIF for AI fingerprints — content credentials, generator metadata and pixel signals — entirely in your browser. Your image never leaves your device.

Drop an image here or click to upload

JPEG · PNG · WebP · GIF · up to 20 MB

Analyzed locally in your browser

ADVANTAGES

Built for fast, defensible answers — not guesses.

Private by default

Detection runs entirely in your browser. Your image is never uploaded, never stored.

Provenance-first

Reads C2PA / Content Credentials and JUMBF blocks used by Adobe Firefly, OpenAI, Microsoft, Leica and more.

Generator fingerprints

Detects Stable Diffusion parameters, ComfyUI workflows, Midjourney, DALL·E, Imagen, FLUX and Firefly signatures.

Camera evidence

Spots Make/Model EXIF and GPS metadata typical of phone and camera photos to balance the verdict.

Pixel heuristics

Lightweight gradient and edge-density checks flag the unnaturally smooth output AI models tend to produce.

Downloadable report

Export a JSON report with verdict, score, signals and metadata — useful for moderation and audit trails.

HOW IT WORKS

Three layers of evidence — combined into one score.

Content credentials

We parse JUMBF / C2PA boxes for cryptographic provenance attached by major AI tools and cameras.

Generator metadata

PNG tEXt / iTXt, JPEG EXIF / XMP and WebP chunks are scanned for SD parameters, ComfyUI graphs, Midjourney tags and IPTC trainedAlgorithmicMedia.

Pixel heuristics

A 256×256 sample is analyzed for gradient and edge density — extreme smoothness leans AI; camera-like noise leans real.

AI detection is heuristic. Strong metadata matches are reliable, but AI images stripped of metadata may still pass — and edited photos can lose their camera EXIF. Use the verdict as a hint, not as proof.

FAQ

Common questions.

Is my image uploaded somewhere?

No. Detection runs entirely in your browser. The image is read and analyzed locally and never sent to a server.

Which generators can it identify?

Stable Diffusion (incl. AUTOMATIC1111 / ComfyUI / InvokeAI / NovelAI / SDXL), Midjourney, DALL·E, Adobe Firefly, Google Imagen, FLUX, Leonardo.AI, plus any image carrying C2PA Content Credentials.

Can it detect AI images with metadata stripped?

It tries — pixel heuristics and common AI output dimensions still give a hint — but stripped images can pass as inconclusive. No client-side tool is 100% accurate.

What does the score mean?

It's an aggregated AI-likelihood (0–100%). 70+ flags AI; 25 or below flags real; in between is inconclusive. Strong metadata signals pin the score high regardless of weak signals.

Is this free?

Yes — detection is free and runs in your browser. No accounts, no subscription.

FINAL CHECK

Confidence in seconds, without uploading a single byte.

Drop an image and get verdict, signals and metadata — useful for journalism, moderation, marketplaces, dating safety and content review.

Open Studio