With Midjourney, DALL·E, Stable Diffusion, and the generators built into ChatGPT and Gemini now producing near-photorealistic images, telling a real photo from a synthetic one has become genuinely hard — sometimes even for a trained eye. This free tool cross-checks several forensic signals to help you make a more informed guess.
👉 Try the AI Image Detector now — free, no sign-up required to run a scan, works entirely in your browser.
What the AI Image Detector Checks
Rather than a single “yes/no” verdict from a black-box AI model, the tool surfaces five independent signals so you can judge the evidence yourself:
| Signal | What it looks at |
|---|---|
| Camera metadata (EXIF) | Whether the file carries device information a real camera normally embeds (make, model, date, settings) |
| Generator fingerprint | Text traces some AI tools leave behind in a file’s metadata (e.g. “Midjourney”, “Stable Diffusion”) |
| Image dimensions | Whether the resolution matches a known standard AI-generator output size (1024×1024, 1792×1024, etc.) |
| Sensor noise residual | Whether the image carries the faint grain a real camera sensor always produces, even in flat areas like sky or skin |
| Error Level Analysis (ELA) | Recompresses the image and compares it to the original to reveal areas with inconsistent compression levels |
Why This Approach, Not Just “AI vs. Human”
The most reliable way to confirm an image’s origin is an invisible watermark embedded at generation time — Google’s SynthID does this, and the C2PA standard attaches cryptographic provenance metadata to the file. But both have a hard limit: they only work if the tool that created the image implements them, and the data survives (a screenshot alone is enough to strip C2PA).
Without a verifiable watermark, forensic analysis is the next best thing — not proof, but a weight of evidence. That’s exactly what this tool does, and it’s upfront about the difference.
Who This Tool Is For
- Online shoppers checking whether a marketplace listing photo is a real product shot or an AI-generated stand-in
- Sellers and buyers on platforms like Vinted or Leboncoin, where AI-generated “damage photos” have been used in refund scams
- Anyone fact-checking an image circulating online before sharing it further
- Content moderators needing a fast first-pass filter before deeper review
Honest Limitations
This detector runs entirely on heuristics computed in your browser — there’s no server, no deep learning model trained on millions of images behind it. That’s a deliberate privacy choice, but it comes with real limits:
- A well-made AI image with clean metadata can trigger zero signals — absence of evidence is not evidence of authenticity.
- A real photo that’s been heavily compressed, cropped, or passed through several apps can trigger false positives.
- It cannot detect SynthID or verify C2PA credentials — for that, the tool links out to the official verification portals.
Use it as a fast first filter, not a courtroom-grade verdict.
AI Detection Glossary
Deepfake
Synthetic content built specifically to impersonate a real person or falsify a real event — a narrower term than “AI-generated image” in general.
SynthID
Google DeepMind’s invisible pixel-level watermark, embedded at generation time. Verifiable only through Google’s own tools (the Gemini app, or the SynthID Detector portal).
C2PA (Content Credentials)
An open, cryptographic provenance standard backed by Google, Adobe, Microsoft, and OpenAI. Attached as metadata — which a screenshot or recompression will typically erase.
ELA (Error Level Analysis)
A forensic recompression technique that highlights zones of an image with an inconsistent compression signature.
Frequently Asked Questions
Does this tool upload my photos anywhere?
No. The entire scan runs in your browser via JavaScript. Nothing is sent to a server, stored, or shared.
Can it detect images made with Gemini or ChatGPT?
Not with certainty. Those platforms increasingly embed SynthID or C2PA credentials, which this tool cannot read — only the platform’s own verification tools can confirm those.
Why did a real photo of mine get flagged?
Usually because it lost its EXIF metadata somewhere along the way — messaging apps and social networks strip it by default. That alone isn’t proof of anything; check the other signals too.
Disclosure: this tool provides indicative forensic signals, not a certified determination of an image’s origin. See the tool page for full method details and limitations.
