gilded-deepfake

byDeepika G

DeepfakeShield — Revised Official Build Prompt 🎯 Core Definition DeepfakeShield is a privacy-first AI web application that detects manipulated images and videos, explains the reasoning, and converts analysis into forensic evidence + real-world response actions. Built for practical AI safety — designed to work reliably in demo conditions. 🏆 Core Features (6 Modules) 🧠 Deepfake Detection Engine Powered by prithivMLmods/Deep-Fake-Detector-Model (HuggingFace transformer) Accepts image uploads; video → auto frame-extract → analyze best frame Outputs: Real / Fake classification + Confidence Score (0–100) Fallback: If model fails → rule-based metadata analysis (file anomaly flags) Zero GPU required, loads in seconds 📊 Trust Score Engine Converts confidence score into user-friendly Trust Score (0–100) Visual Risk Indicator: 🟢 Safe / 🟡 Suspicious / 🔴 Fake Detected Formula: Trust Score = 100 - (fake_confidence * 100) 📄 Evidence Report Generator (Forensic-Grade) Downloadable PDF containing: Detection Result + Trust Score + Confidence Score Timestamp + SHA-256 File Hash QR Code linking to verification summary Short Analysis Summary Libraries: ReportLab + qrcode + hashlib 🧠 Explainable AI Output Text-based explanation of WHY the model flagged it Highlights: lighting inconsistency / facial blending / compression artifacts No Grad-CAM (time risk) — clean text output is enough for judges 🚨 Response Mode System Triggers when: Fake detected OR Trust Score < 40 Provides: Alert Screen + Risk Warning One-click: Download Report + Copy Complaint Text Platform reporting guide (Instagram / YouTube / WhatsApp) Auto-generated complaint text aligned with Indian Cybercrime Portal (cybercrime.gov.in) format 🔐 Privacy-First Architecture Files processed in-memory only (tempfile + auto delete) No cloud storage, no database Only metadata retained: hash + result + score (session only) 🔄 System Workflow Upload Image/Video ↓ SHA-256 Hash Generated ↓ Pre-processing (frame extract if video) ↓ HuggingFace Detector runs ↓ Trust Score calculated ↓ Text-based XAI explanation generated ↓ Evidence PDF created (hash + QR + result) ↓ If high risk → Response Mode activates ↓ User downloads report / copies complaint ↓ File deleted from memory 🏗️ Tech Stack LayerTechnologyFrontendReact + Tailwind CSSBackendPython + FlaskAI ModelHuggingFace Transformer (Deep-Fake-Detector)Video ProcessingOpenCV (frame extraction only)PDF GenerationReportLab + qrcodeSecurityhashlib SHA-256File Handlingtempfile (in-memory, auto-delete) 📁 Folder Structure deepfakeshield/ ├── frontend/ │ ├── src/ │ │ ├── components/ │ │ │ ├── UploadZone.jsx │ │ │ ├── ResultCard.jsx │ │ │ ├── TrustScore.jsx │ │ │ ├── ResponseMode.jsx │ │ │ └── ReportDownload.jsx │ │ └── App.jsx ├── backend/ │ ├── app.py ← Flask API │ ├── detector.py ← HuggingFace model │ ├── trust_score.py ← Score calculator │ ├── report_gen.py ← PDF + QR + Hash │ ├── explainer.py ← Text XAI │ └── response_mode.py ← Complaint generator ├── requirements.txt └── README.md ⏱️ 24-Hour Build Timeline Time BlockTask0–1 hrSetup repo, install dependencies, test HuggingFace model loads1–3 hrdetector.py + trust_score.py + Flask /analyze API endpoint3–6 hrreport_gen.py — PDF + SHA-256 + QR code working6–9 hrReact frontend — Upload + Result + TrustScore UI9–11 hrresponse_mode.py + Complaint text generator11–12 hrexplainer.py — text XAI output12–13 hrConnect frontend ↔ backend fully13–15 hrFull flow test (image + video both)15–16 hrBug fixes + edge case handling16–17 hrUI polish + mobile responsiveness17–18 hrDemo script practice + README 🎤 Judge Pitch (2 Minutes) Problem: "Deepfakes are destroying trust — and victims don't know what to do" Solution: "Upload any file → instant detection → forensic PDF → cybercrime complaint ready" Demo: Upload → Trust Score → PDF download → Response Mode → Complaint text copy Close: "Privacy-first, works offline, built for Indian users facing cyber harassment" 🏆 Your Real Differentiators PDF with QR + SHA-256 — no other team will have this Indian Cybercrime Portal complaint generator — hyper-local, judges will love it Privacy-first in-memory processing — ethical AI angle Response Mode — converts AI output into real action 1. UI/UX DESIGN (VERY HIGH PRIORITY) - Minimal, clean, futuristic UI - Apple-level spacing and typography - Smooth animations (micro-interactions everywhere) - Dark + Light mode support - Glassmorphism + soft shadows (not overused) - Consistent design system (colors, spacing, buttons, cards) - Mobile-first responsive design - Professional dashboard layout 2. INFORMATION ARCHITECTURE - Very clear navigation (no confusion) - Sidebar + topbar layout (if dashboard) - Logical grouping of features - User flow must be simple (max 3 clicks to any feature) 3. FEATURES (MUST BE POWERFUL BUT PRACTICAL) - AI-powered core feature (analysis / prediction / assistance depending on app type) - Real-time data updates - User authentication system (login/signup) - Profile management - History tracking (logs / activity / results) - Export or download feature (PDF/Report generation) - Smart notifications system 4. DASHBOARD DESIGN - Central dashboard with summary cards - Analytics charts (clean, not cluttered) - Quick action buttons - Recent activity feed - AI insights panel 5. USER EXPERIENCE RULES - No clutter, only meaningful elements - Every screen must have a purpose - Empty states must be beautifully designed - Loading states must be animated - Error handling must be user-friendly 6. AI INTEGRATION - AI should explain results in human-friendly language - Provide “Why this result?” section - Suggest next actions automatically - Provide confidence score or reasoning indicators 7. OUTPUT REQUIRED Generate: - Full UI layout structure (screen-by-screen) - Component breakdown - Feature list (detailed) - User flow diagram explanation - Color palette + typography suggestion - Sample dashboard design description - Mobile + desktop adaptation 8. FINAL GOAL The product must feel like: "A startup-ready SaaS product that can be launched immediately and impress judges in a hackathon." ──────────────────────────── 🚀 IMPORTANT RULE ──────────────────────────── Do NOT create a basic app. Avoid generic templates. Make it feel premium, intelligent, and investor-ready. Focus on clarity, speed, usability, and visual excellence. create a app with 99% accuracy