DEV Community

Cover image for I got tired of guessing if "that" image was real. So I built a tool to prove it.
Gaurav Pandey
Gaurav Pandey

Posted on

I got tired of guessing if "that" image was real. So I built a tool to prove it.

We’ve all had that moment. You’re scrolling, you stop at a photo, and something feels... off. The lighting is too perfect. The skin texture is too smooth. But you can't be sure.

We are rapidly losing our ability to trust what we see on the web.

Existing detection tools are often "black boxes"—they give you a random percentage without explaining why. I didn't want a guess; I wanted evidence.

So I spent some time building AI Content Scanner—an open-source Chrome extension that doesn't just look at the pixels; it digs into the digital DNA of the content you consume.

Here is the breakdown of the build:

1. The Verdict (No Magic, Just Math)

The extension assigns a "Confidence Score" (0–100%) to every image, video, and text block. But unlike other tools, it tells you exactly how it got there.

  • Likely Real: No digital fingerprints found.
  • AI Detected: Hard evidence found (metadata tags, signatures).
  • Uncertain: The file is messy, but has suspicious traits.

2. The Granular Details (How it works)

I didn’t want to rely on visual classifiers that hallucinate. I focused on deterministic signals—the stuff the models leave behind.

  • C2PA & JUMBF: It scans the binary data for the new cryptographic provenance standards used by responsible AI generators.
  • Metadata Forensics: It parses EXIF and XMP data looking for specific signatures from tools like Adobe Firefly, DALL-E 3, and Midjourney.
  • Google SynthID: It checks for the specific watermarking techniques used by Imagen and Gemini.
  • LLM "Slop" Detection: For text, it analyzes sentence uniformity and "transition word density." (If a paragraph uses the word "delve" or "tapestry" three times, it’s probably GPT-4).

3. The Show (The Experience)

I wanted it to be passive. You don't have to upload anything. You just browse the web. When the extension detects something synthetic, it slaps a badge on it. You hover over the badge, and it shows you the raw data: "Source: Midjourney v6" or "Software: Adobe Firefly."

It is transparent, it is local, and it is open source.

If you are a dev, I’d love for you to poke around the code, break it, or add new signatures.

👉 Repo & Download: https://github.com/gauravkrp/ai-content-scanner
👉 Website: https://ai-content-scanner.vercel.app


⚠️ Note on Installation:
The extension is currently under review by the Chrome Web Store (pending approval). But since it's open source, you don't have to wait to use it:

  1. Go to the Releases section on the GitHub repo and download extension.zip.
  2. Open chrome://extensions in your browser.
  3. Toggle "Developer mode" on (top right corner).
  4. Drag and drop the zip file into the window.

(It takes about 10 seconds, and you’re good to go.)

#OpenSource #Engineering #WebDev #AI #Transparency #ChromeExtension #BuildInPublic

Top comments (0)