DEV Community

Cover image for I Built an AI Fake Image Detector in a Weekend — And 100 People Used It on Day 1
Viraj Parab
Viraj Parab

Posted on

I Built an AI Fake Image Detector in a Weekend — And 100 People Used It on Day 1

#ai

Last weekend I decided to experiment with a small idea:
What if there was a quick, developer-friendly way to check whether an image is AI-generated, edited, or unsafe?

There are big companies working on moderation tech, but most of their tools are either:

hidden behind heavy enterprise sales funnels,

bundled with extra features you don’t need, or

not accessible to everyday devs building side projects.

So I built a lightweight version:
👉 https://authenticimage.site/

(No logins, no paywalls — just upload an image and get the result.)

To my surprise, ~100 people used it within the first day, without me promoting it anywhere. That unexpected traction made me realize there’s genuine curiosity — and maybe even real need — for image authenticity tooling.

⚙️ How it Works (High-Level Architecture)

I combined a few open-source and custom models into a single detection pipeline:

  1. AI / GAN Detection

CNN-based classifier

Frequency analysis

GAN fingerprint models

Artifact + noise pattern analysis

This catches images from models like Stable Diffusion, Midjourney, etc.

  1. Image Manipulation Detection

Error Level Analysis (ELA)

Noise inconsistency checks

Splicing/clone detection

Metadata extraction (EXIF anomalies)

Helpful for spotting tampering or edits.

  1. Safety / Moderation

Nudity

Gore/violence

NSFW

Unsafe content

Deepfake indicators

These run lightweight and return results in ~50–100ms depending on the payload.

🧪 What I Learned From the First 100 Users

Even with no promotion, users found it and tested:

their AI art

dating profile pictures

stock image samples

edited meme formats

social media photos

suspicious images from group chats

The variety surprised me. I still don’t know exactly which audience I should build for, so for now the tool remains completely free while I gather usage signals.

🧭 What I’m Trying to Figure Out Next

I'd love feedback from the DEV community on:

Should I turn this into an API for developers?

What accuracy threshold do you expect from an AI image detector?

What features are missing? (Explainability? Logs? Batch verification?)

Would you use this in a personal or professional project?

And if you want to play with the tool:
👉 https://authenticimage.site/

💬 Final Thoughts

This was one of those projects where I didn’t overthink it — I just built it because I was curious. Seeing real users show up without any promotion made me realize there’s more demand for digital authenticity tools than I expected.

If enough devs find it useful, I’m happy to:

open-source some components,

publish the full pipeline architecture,

create a public API,

or grow it into something bigger.

Let me know what you think, and feel free to break it — the more feedback I get, the better it becomes!

Happy building 🚀

Top comments (0)