VibeVerify AI

VibeVerify AI: Instant deepfake detector with fun personality roasts spot AI-polished profiles in seconds, get your celeb twin & dateability score. Built for 2026 trust crisis, addictive & shareable

Comment

VibeVerify AI: The Fun Deepfake Detector That's Actually Addictive

Inspiration

In the last 5 months (Oct 2025–Feb 2026), deepfake fraud has exploded globally: $25M Arup CEO video scam (KnowBe4 Feb 2026), billions lost in romance/investment scams (Resemble AI Q3 2025, WEF). Everyone's terrified to trust online—dating feels broken, video calls suspicious, social media full of fakes.

Inspired by emerging platform tools (TikTok AI flags, LinkedIn profile warnings), I built VibeVerify as the first personal "reality layer" for 2026. It's not boring detection—it's a gamified, hilarious personality roast that makes people want to check pics voluntarily. Global scope: solves trust erosion everywhere, no Africa limitation.

What It Does

Upload any photo/screenshot → <10 seconds later:

  • Realness Score: 92% REAL 🔥 or 58% AI POLISHED 👀 (catches metadata like C2PA + visual fakes).
  • Viral Personality Report: Celeb twin (e.g., Idris Elba vibes), dateability %, meme energy, archetype—pure Gen-Z fun.
  • Gamification: Daily streak, leaderboard, shareable vibe cards (viral by design).

Users spam it on selfies, crush pics, group shots—just for the laughs. But under the hood, it protects against scams.

How I Built It

Started with the hackathon protocol: Hook to 2026 deepfake crisis → Pivot to fun UX → Tech core.

  • Detection Engine: Hugging Face Inference (dima806/ai_vs_real_image_detection model for visuals + BLIP captioning for identity analysis). Custom metadata scanner reads file bytes for C2PA/AI signatures (DALL·E, Midjourney, Adobe AI)—catches polished headshots like LinkedIn.
  • Personality Roast: Groq (Llama-3.3-70B) generates savage, demographic-smart reports (e.g., accurate celeb twins for Black professionals in suits).
  • Frontend: Next.js 15 (App Router), TypeScript, Tailwind CSS for mobile-first, single-screen UX. Lucide icons, Sonner toasts, canvas-confetti for fireworks.
  • Gamification: LocalStorage for streaks, fake leaderboard modal.
  • DevOps: GitHub repo with CI/CD, Vercel free tier deploy. 100% free APIs—no paid tiers.

MVP in <24 hours: Feasibility audit passed (HF/Groq free, Vercel deploy). No complex dashboards—simple, lovable, gamified.

Challenges I Faced

  • Accuracy for Polished AI: Free models miss subtle edits (suit swaps, fake wrinkles). Solved with metadata scanner + threshold tuning for "gray zone" (60-85% = polished). Tested on my own AI-edited headshot—now flags it perfectly.
  • Fun Factor: Detection alone is boring. Iterated Groq prompts for Gen-Z roasts—now addictive (people share for celeb twins/meme energy).
  • UX Magic: Enforced <10s "aha!"—optimized HF calls with retries, added score animation/confetti.
  • Hackathon Constraints: 100% free tiers, GitHub repo, live Vercel URL—demo or die.

Learned: Multimodal AI (visual + metadata) is key for 2026 deepfakes. Next.js makes MVPs fly.

What I Learned & Future Plans

Learned advanced AI inference (HF ensembles, metadata forensics) + viral UX design. Future: Add voice/video analysis, B2B for dating apps (premium checks), mobile app. Million-dollar potential: Freemium model with pro roasts.

VibeVerify isn't just safety it's the fun trust tool 2026 needs. Try it and get roasted!

Built With

Share this project:

Updates