Free tool · AI & manipulation check

Is this image AI-generated?

Capture it · Seal it · Verify it

Run a free check below — it works entirely in your browser, so your image is never uploaded. It reports an open-source detector's score and any Content Credentials (C2PA) the file carries. We're honest about the limits: no algorithm can prove with certainty whether an image is AI-generated or manipulated.

Try it now, no account needed.Your first proof is free · €10 welcome credit when you confirm your email · No card neededLaunch offer. Credit can also be used for webpage captures or file certification.
In-browserimage never uploaded
C2PAauthoritative when present
Honestindicative, not proof
AI image check illustration
  1. 01

    Capture it

    Save a webpage, record a browser session, or certify a file.

  2. 02

    Seal it

    A cryptographic seal and timestamp help reveal later changes.

  3. 03

    Verify it

    Anyone can check the evidence package independently.

Free AI-image check.

Runs on your device with on-device machine learning. Your image never leaves your browser — we don't see it or store it.

Read this first. A detector score is an indication, not proof. Independent benchmarks find open-source detectors are right only ~50–65% of the time on modern AI images, and the newest generators evade them. No method can establish with certainty that an image is — or is not — AI-generated or manipulated. Content Credentials (C2PA) are authoritative when present; their absence proves nothing.

Why no tool can be certain.

We'd rather tell you the truth than sell you a false yes/no.

~
Detectors are a coin-flip on new generatorsBenchmarks show 50–65% accuracy out-of-distribution, and 18–30% on the latest models (Flux, Firefly, Midjourney v7). High "in-distribution" scores don't transfer to the real world.
↔
It's an arms raceEvery new image model is trained to look more real; detectors trained on yesterday's models miss tomorrow's. A confident-looking score can be confidently wrong.
✓
Only cryptographic provenance is authoritativeContent Credentials (C2PA) are signed at creation by some cameras and AI tools — when present they don't guess, they prove. But most images carry none.

What the free check looks at.

AI?

Open-source detector score (indicative)

A named open model runs on your device and returns a likelihood that the image is AI-generated. Useful as a hint — never as proof.

C2PA

Content Credentials (authoritative)

We read any signed C2PA provenance in the file — e.g. "generated by DALL·E" or "edited in Photoshop". More on verifiable records →

🔒

100% local

Both checks run in your browser. The image is never uploaded — maximum privacy, and nothing for us to store.

Need a record that holds up?

The free check tells you what today's tools say — a hint, not proof. What InstantProof can prove is that you held this exact image, unchanged, at a given date: certify the file itself.

📄
Certify the image itselfThe file is SHA-256 hashed, stamped with a qualified RFC 3161 timestamp and signed — a tamper-evident record that this exact image existed, unchanged, on that date.
🕓
Proof of possession & priorityEstablishing that you held the image, at a specific moment, is what carries weight in IP, copyright and authenticity disputes — independently of the AI question.
🌐
Captured at the source, not screenshottedIf the image is on a website or in a chat, a certified capture from the real server proves it was published there, and when. A screenshot or re-upload cannot. →
Certify a file Capture a web page or chat

AI-image-check questions.

Can a tool tell for sure if an image is AI-generated?

No. No algorithm can determine with certainty whether an image is AI-generated or manipulated. Independent benchmarks show open-source detectors are right only about 50–65% of the time on modern, out-of-distribution generators — and the newest models evade most detectors. A detector score is an indication, never proof. The one authoritative signal is Content Credentials (C2PA): cryptographically-signed provenance that some cameras and AI tools embed — but most images don't carry it.

Does my image get uploaded?

No. This check runs entirely in your browser using on-device machine learning. The image never leaves your computer and we never see or store it.

What are Content Credentials (C2PA)?

A cryptographically-signed record of how a file was created and edited, based on the open C2PA standard. Adobe, Microsoft, OpenAI, Leica, Nikon and others embed them. When present they are authoritative; when absent, nothing can be concluded from their absence.

Is the result legal proof?

No — the free check is indicative only. What InstantProof can certify is the file itself: a SHA-256 hash sealed with a qualified RFC 3161 timestamp and signature, proving you held this exact image, unchanged, at a given date. That's a tamper-evident record of possession and priority — it doesn't turn an uncertain detector score into a certainty.

Is this legal advice?

No. This service creates technical evidence artifacts. Legal admissibility depends on jurisdiction and circumstances. Consult qualified legal counsel for advice specific to your situation.

Digital evidence. Made simple.

Capture what matters. Before it changes.

Start with €10 welcome credit. Keep a cryptographically sealed evidence package you can download and verify.

Capture your first proof →