Skip to content
perspicaxby avancer.ai (opens in a new tab)

AI Image DetectorKnow when a photo was made by AI.

A free AI image checker from the avancer.ai research lab. Perspicax reads an image's Content Credentials first, then its pixels, and answers in seconds: likely AI-generated, likely real, or inconclusive.

Paste an image or a link anywhere (Ctrl/⌘ + V)

Your first scan is free. By scanning, you accept our Terms. Images and results are kept until you delete them (Privacy).

  • No account needed
  • Free account: 2 scans a day
  • Fewer than 1 in 100 camera photos flagged
Illustration of a scan. The photo was generated with Nano Banana Pro.
In short

What is Perspicax?

Perspicax is an AI image detector. Drop or paste a photo, and TrueEye 2.1 checks its Content Credentials (C2PA) and its pixels, then answers likely AI-generated, likely real or inconclusive. In testing it caught up to 99% of AI images, depending on the model, and flagged fewer than 1 in 100 camera photos.

Quick check: Is this real or AI? · Guide: How to tell if an image is AI generated

Accuracy

Up to 99%of AI images caught, depending on the model
  • 92.1–93.9%of ChatGPT images (GPT Image) caught
  • 93.6–98.3%of Gemini images (Nano Banana, Imagen) caught
  • 91.0–98.9%of Seedream images caught
  • 0.6%of original camera photos wrongly flagged: fewer than 1 in 100 (99.4% called real)
See all results
Why labels are not enough

We made this photo with AI. One re-save erased the label.

Google's image model signs its files with Content Credentials (C2PA). We generated this market photo, saved it once for the web, and the signature was gone. Screenshots and many social platforms do the same. Perspicax reads the label when it is there, and the pixels when it is not.

Provenance checks
  • Content Credentials (C2PA)Live
  • SynthID watermark checkSupported
A spice seller in an old market at golden hour. The image was generated by AI.
ORIGINAL FILE FROM GOOGLE'S MODEL“Created by Google Generative AI”Signed by Google C2PA Root CA G3, with a SynthID watermark noted.
SAME PHOTO, SAVED ONCENo Content CredentialsNothing in the file says AI made it. Only the pixels can.
Your turn

Real or AI?

Six photos a newsroom, a marketplace or a delivery app could receive today. Make your call on each, then reveal.

  • A young woman laughing by a cafe window
    Cafe portrait
  • A tray of kunafa with a slice being lifted
    Restaurant dish
  • A pair of worn white sneakers on a balcony floor
    Marketplace listing
  • A golden retriever sitting on grass in a park
    Phone snapshot
  • Aerial view of red sand dunes with a white car
    Drone photo
  • Hands pouring Arabic coffee from a brass pot
    Hospitality
0 of 6 answered
How it works

Labels first. Then every pixel.

How Perspicax sees an imageTrueEye reads an image region by region, down to the pixels. Move your pointer to tilt it.

  1. Read the label

    If the file still has Content Credentials (C2PA) or an IPTC source tag, we show them first.

  2. Check it is a photo

    Screenshots, documents and graphics come back as outside scope instead of a guess.

  3. Read the pixels

    TrueEye 2.1 examines the whole image, region by region, down to the pixels.

  4. Answer in three bands

    Likely AI-generated, inconclusive or likely real. When the evidence is not strong enough, we say so.

See how it works
What it detects

Built for the images people actually see.

Photos from the apps and models in wide use today, measured one by one.

AI images caught by Perspicax in testing, measured model by model
Images fromCaught in testing
ChatGPT (GPT Image)GPT Image 2.5 Sunburst, GPT Image 2.5 Flare, GPT Image 292.1–93.9%
Gemini (Nano Banana, Imagen)Imagen 4, Nano Banana 2.1, Nano Banana Pro93.6–98.3%
Seedream (Doubao, Dreamina, CapCut)Seedream 4.5, Seedream 5.0 Pro, Seedream 5.0 Lite91.0–98.9%
Grok ImagineGrok Imagine 2.096.2%
FLUXFLUX 396.6%
Other new modelsHunyuan 3.0, HiDream-O1, MAI-Image 2.5, Qwen-Image 3.0up to 99.9%
Original camera photos99.4% called real
All generators
Benchmarks

Accuracy you can check.

Measured on public test sets at a decision point of p = 0.5.

  • 99.4%Camera photosOriginal camera photos correctly called real.
  • 98.0%AIGenImages2026A 2026 benchmark of real and AI images. AUC 99.7.
  • 97.8%ImageTimeAI images caught.
  • 97.1%OpenFake, RedditReal and AI images posted on Reddit.
  • 95.6%So-Fake-OODReal and AI images from social media.
  • 95.1%OpenFakeIts core set of real and AI still images.
  • 93.7%DailyBenchSeven current generators.
  • 93.5%RealChainImages passed along real sharing chains.
  • 91.8%Never-seen generators17 generators the model never saw in training, 31,334 images.
Open the accuracy page
Who it's for

Who uses an AI image detector

  • Newsrooms and fact-checkers

    Check a photo before you publish it.

  • Trust and safety teams

    Screen uploads and profile photos at scale.

  • Marketplaces

    Catch AI-made listing photos before they go live.

  • Everyone else

    Check a viral photo before you share it.

For developers

One request. One verdict.

Wait for the answer in the same call, or take a 202 and poll. Idempotency keys, clear errors, and the model version on every result.

Anyone can request access. Enabled within an hour.

Request API accessRead the API guide
POST /api/v1/scans · example
$ curl https://perspicax.io/api/v1/scans \    -H "Authorization: Bearer $PERSPICAX_KEY" \    -H "Prefer: wait=10" \    -F [email protected]# 200 OK{  "status": "completed",  "result": {    "label": "ai_generated",    "score": 0.97,    "provenance": { "c2pa": null }  },  "model": { "name": "TrueEye", "version": "2.1" }}
On the roadmap

What's coming next.

Pricing

Free to start. Simple to grow.

No account yet? Your first scan is free.

  • Free

    $0No card needed. Sign in with just your email.

    • 2 scans a day per member
    • Verdict, Content Credentials and metadata
    • SynthID watermark check
    • History kept forever
    • Dispute a verdict
    Create a free account

  • Pro

    10× Free

    $20a month ($200 a year, two months free)

    • 20 scans a day, for one person
    • Everything in Free
    • Region-by-region detail for every scan
    • History kept forever
    Upgrade to Pro

  • Team

    Per member

    $25per member per month ($250 a year)

    • 20 scans a day per member
    • Invite your team
    • Separate teams and subscriptions
    • Everything in Pro
    Start a team

  • Enterprise

    CustomCredits for your team

    • Credits that refresh monthly or yearly
    • Top-ups any time, alerts before you run out
    • Live usage dashboard
    • API keys and retention options
    Contact us
Compare plans

AI image detector questions

What is an AI image detector?

A tool that tells you whether a picture was made by an AI image generator. Perspicax checks photo-like images: it reads any Content Credentials, examines the pixels and answers in three bands, with inconclusive when the evidence is not strong enough.

Which AI generators can Perspicax detect?

It is built for AI-generated photos in general. We publish measured results for ChatGPT (GPT Image), Gemini (Nano Banana, Imagen), Seedream, Grok Imagine, FLUX and other new models.

Is the AI image checker free?

Yes, to start: your first scan needs no account, and a free account gives you 2 scans a day. Pro and Team add more.

What happens to my images?

They stay in your scan history until you delete them, and they are not used to train TrueEye 2.1 unless you opt in.

Does it work on edited photos or video?

No. Perspicax checks still images that were fully generated by AI. Edits to real photos and video are outside its scope.

Before you believe it, check it.

Your first scan is free, no account needed. A free account gives you 2 scans every day.

Scan an image