I Built an AI Image Detector - Here Is What I Learned

The Problem

Every day millions of AI images flood the internet. Most people cannot tell the difference. I built a tool to help.

What I Built

TruthLens - an AI image detector. Upload any image, get instant verdict: real or AI-generated.

Tech Stack

• Next.js 14 App Router
• Vercel serverless
• Supabase PostgreSQL
• Stripe billing
• Multi-layer AI detection

How Detection Works

Layer 1 - Pixel Analysis: AI generators leave noise patterns that differ from cameras. FFT analysis reveals these.

Layer 2 - Semantic Analysis: Trained models spot lighting, shadow, and reflection inconsistencies.

Layer 3 - Metadata: Real photos have EXIF data. AI images dont.

Business Model

• Free: 10 scans/day
• Pro: $9.99/month, 100 scans/day
• Lifetime: $199 one-time
• Enterprise: $299/month
• API: Developer access

What I Learned

1. Distribution harder than building
2. SEO takes months - published 9 blog articles
3. AI directories drive traffic
4. Freemium builds trust
5. Dev.to is underrated for developer products

Try It

TruthLens - no signup needed.

GitHub

Feedback welcome!