8080.AI
DesignReqsTasksUser FlowArchitecture
Log inSign up for free

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

No preview

Comments (0)

No comments yet. Be the first!

System Requirements

System Requirement Document

System Requirements Document (SRD)

Project Name: gilded-deepfake

1. Introduction

The gilded-deepfake project, spearheaded by Deepika G from India, is a privacy-first AI web application designed to detect manipulated images and videos, explain the reasoning behind the detection, and convert the analysis into forensic-grade evidence. This application is tailored to address the growing concerns surrounding deepfakes, providing users with actionable tools to combat cyber harassment and misinformation. The system is optimized for Indian users, incorporating hyper-localized features such as integration with the Indian Cybercrime Portal.

Page 1 of 7

2. System Overview

gilded-deepfake is built to deliver reliable deepfake detection and analysis in demo conditions, ensuring high accuracy and usability. The system processes files entirely in-memory, maintaining user privacy by avoiding cloud storage. Key functionalities include:

  • Deepfake Detection Engine: Analyzes images and videos using advanced AI models.
  • Trust Score Engine: Converts detection confidence into a user-friendly trust score.
  • Evidence Report Generator: Creates forensic-grade PDFs with detection results, timestamps, and QR codes.
  • Explainable AI Output: Provides text-based explanations for detection results.
  • Response Mode System: Generates actionable steps for users, including complaint text aligned with Indian cybercrime reporting standards.

The system workflow is streamlined to ensure minimal user effort while delivering maximum impact.

3. Functional Requirements

Page 2 of 7

As User:

  • I should be able to upload images or videos for analysis.
  • I should receive a trust score and detection result for uploaded files.
  • I should be able to download a forensic-grade evidence report in PDF format.
  • I should see a text-based explanation of why a file was flagged as fake.
  • I should be alerted when a file is classified as fake or has a trust score below 40.
  • I should be able to copy pre-generated complaint text for reporting to the Indian Cybercrime Portal.
  • I should be guided on how to report flagged content on platforms like Instagram, YouTube, and WhatsApp.

As Admin:

  • I should be able to monitor system performance and logs.
  • I should be able to update AI models and detection rules.
  • I should ensure compliance with privacy standards and manage session-based metadata.

4. User Personas

Persona 1: Individual User

  • Description: Victims of cyber harassment or misinformation seeking tools to verify and report manipulated content.
  • Goals: Detect deepfakes, generate evidence, and take action against cyber threats.
  • Pain Points: Lack of technical expertise, difficulty in reporting incidents.
Page 3 of 7

Persona 2: Legal Professionals

  • Description: Lawyers or forensic experts requiring reliable evidence for legal cases.
  • Goals: Obtain forensic-grade reports and explanations for court submissions.
  • Pain Points: Need for high accuracy and privacy in processing.

Persona 3: Content Moderators

  • Description: Moderators from social media platforms or organizations aiming to maintain content integrity.
  • Goals: Quickly detect and flag manipulated content.
  • Pain Points: Scalability and speed of detection.

5. Visuals Colors and Theme

Color Palette:

  • Background: #1E1E2F (Deep Midnight Blue)
  • Surface: #2A2A40 (Charcoal Gray)
  • Text: #EAEAEA (Soft White)
  • Accent: #FF6F61 (Vivid Coral)
  • Muted Tones: #A1A1B5 (Muted Lavender)

Theme:

The design will feature a clean, futuristic aesthetic with glassmorphism elements, soft shadows, and smooth animations. Both dark and light modes will be supported, ensuring accessibility and user preference.

Page 4 of 7

6. Signature Design Concept

Interactive Evidence Dashboard

The homepage will feature an Interactive Evidence Dashboard that visually represents the uploaded file's analysis journey. Users will see:

  1. Dynamic Flow Animation: A visual pipeline where the file moves through stages like hashing, AI detection, trust score calculation, and report generation. Each stage lights up as it completes.
  2. Trust Score Gauge: A circular gauge with animated transitions, shifting colors from green to red based on the score.
  3. 3D File Preview: A rotating 3D thumbnail of the uploaded file, with overlays showing detected anomalies.
  4. Micro-Interactions: Hover effects on buttons and cards, with subtle vibrations and color shifts.
  5. QR Code Reveal: A glowing QR code that appears with a "scan me" animation when the report is ready.

This concept ensures users feel engaged and informed throughout the process, making the experience unforgettable.

7. Non-Functional Requirements

  • Performance: The system must process files within 5 seconds for images and 10 seconds for videos.
  • Scalability: Support up to 100 concurrent users in demo conditions.
  • Security: Ensure all files are processed in-memory and deleted post-session.
  • Localization: Provide complaint text and platform guides tailored to Indian users.
  • Accessibility: Ensure compatibility with screen readers and keyboard navigation.
Page 5 of 7

8. Tech Stack

Frontend:

  • React for Web
  • Tailwind CSS for styling

Backend:

  • Python
  • FastAPI for API development

Database:

  • No database for file storage; session-based metadata only.

AI Models:

  • HuggingFace Transformer (Deep-Fake-Detector)
  • Google Nano Banana for image generation

AI Tools:

  • Langchain for workflow orchestration
  • Litellm for LLM routing

Local Orchestration:

  • Docker
  • docker-compose
Page 6 of 7

Server-Side Orchestration:

  • Kubernetes for scalability

9. Assumptions and Constraints

  • Files will not exceed 100 MB for videos and 10 MB for images.
  • The system will operate in demo conditions with limited concurrent users.
  • No permanent storage of user files or results.
  • The application will prioritize Indian users, with complaint text aligned to local cybercrime reporting standards.

10. Glossary

  • Deepfake: AI-generated manipulated media that mimics real images or videos.
  • Trust Score: A user-friendly representation of the confidence in detection results.
  • Explainable AI (XAI): AI outputs that provide reasoning behind decisions.
  • SHA-256: A cryptographic hash function used for file integrity verification.
  • Glassmorphism: A design trend featuring translucent surfaces and soft shadows.

This SRD ensures gilded-deepfake is built to deliver an impactful, privacy-first solution for detecting and combating deepfakes. Let’s bring this vision to life!

Page 7 of 7

No completed page designs yet.

Completed design pages will appear here when they are ready to preview.

No user flows yet.

The User Flow Agent will generate per-persona navigation diagrams after SRD updates.

No completed page designs yet.

Completed design pages will appear here when they are ready to preview.

No user flows yet.

The User Flow Agent will generate per-persona navigation diagrams after SRD updates.