Is this image AI-generated?
Upload any photo. Cautellus runs four detection layers in parallel — latent-space artifacts, frequency signatures, metadata inspection, and anatomy checks — and tells you which generative model (if any) made it.
Four layers, run in parallel.
Latent-space artifacts
Diffusion models leave statistical fingerprints in pixel-noise distribution that human eyes can't see. Cautellus extracts those distributions and compares them against signatures from Midjourney, DALL·E, Stable Diffusion, and Flux.
Frequency-domain signatures
Fourier-transform analysis surfaces the regular grid patterns produced by neural upscaling, GAN convolutions, and diffusion sampling — invisible in the spatial domain.
Metadata + EXIF inspection
Real cameras stamp images with sensor metadata: ISO, focal length, GPS, color profile. AI-generated images are missing this — or have suspiciously perfect synthetic metadata that gives them away.
Anatomy + lighting consistency
Hands with seven fingers. Eyes that catch light from impossible directions. Earrings that don't match. Cautellus checks dozens of "tell" categories that current models still get wrong.
Trained against the current generation.
Signatures refresh when new public model versions ship. Detection lags by days, not months.
Where AI images are weaponized.
Dating-app profile photos
Pig-butchering and romance scams now use AI faces that pass casual inspection. Run the profile pic through Cautellus before you trust the face.
Marketplace product listings
Fake online stores generate product photography from scratch. Detection catches the diffusion patterns scammers can't scrub.
Influencer / celebrity endorsements
Crypto and supplement scams paste real celebrity faces onto fabricated product shots. The artifact signature gives them away.
News and social-media images
Misinformation rides on AI-generated "photos" of events that never happened. Verify before you share.
Job-offer ID photos
Fake recruiters request your driver's license photo, then mix it with AI-generated docs. Same detection catches the synthetic side.
Real-estate / rental listings
Vacation-rental scams use AI-staged interiors. The lighting and corner geometry usually fail the diffusion check.
increase in AI-enabled fraud, 2024 → 2025
projected AI fraud losses by 2027
of new internet images projected synthetic by 2026
people approached by AI on a dating app
Frequently asked.
How accurate are AI image detectors in 2026?
No detector is perfect — image generators evolve faster than detection models. Cautellus runs four layers in parallel which catches most unedited model outputs. Heavy editing is harder for any detector to catch.
Can it spot Midjourney v7, SD3, or Flux outputs?
Yes for the current public model versions. Signatures are updated when new models ship.
What about partial AI edits (face swap, inpainting)?
Cautellus flags regions showing generative artifacts even when the rest is real — so a real photo with an AI-swapped face comes back as "AI elements detected" with a heatmap.
Will it ever say "AI" when an image is real?
False positives happen on heavily filtered, AI-upscaled, or aggressively recompressed real photos. The detector returns a confidence score, not a binary verdict.
Is it free?
Your first scan is free, no account needed. Unlimited scans require Cautellus Plus at $9.99/mo after a 7-day free trial that requires a card and auto-renews unless cancelled.
One upload. Four layers. Seconds.
Don't trust the photo until you've checked the photo.
Run the detector