The AI Forensics Handbook

Detection and Analysis of Artificially Generated Text, Images, Video, and Audio

Introduction

Chapter 1: Generative AI Foundations

1. Probability Distributions and Sampling
2. Likelihood, Loss Functions, and Optimization
3. The Generation Pipeline from Model to Output
4. Where Artifacts Originate in Generative Models
5. Model Capacity, Overfitting, and Memorization
6. Transfer Learning and Its Forensic Implications

Chapter 2: Language Models and Text Generation

1. Transformer Architecture: Attention, Layers, and Representations
2. Tokenization, Vocabulary, and Decoding Strategies
3. Next-Token Prediction and Probability Distributions
4. Temperature, Top-k, Top-p, and Their Forensic Consequences
5. Emergent Properties and Capability Scaling
6. Fine-Tuning, Prompting, and Behavioral Variation

Chapter 3: Generative Image Models

1. Generative Adversarial Networks: Training Dynamics and Artifacts
2. Variational Autoencoders and Latent Representations
3. Diffusion Models: Forward Process, Reverse Process, and Sampling
4. Score-Based Models and Noise Scheduling
5. Architectural Fingerprints and Latent-Space Structure
6. Image Quality Metrics and Their Forensic Relevance

Chapter 4: Audio and Speech Synthesis

1. Audio Representation: Waveforms, Spectrograms, MFCCs
2. Waveform Synthesis: WaveNet, WaveRNN, and Autoregressive Models
3. Vocoder Architectures: Griffin-Lim, Neural Vocoders, HiFi-GAN
4. Text-to-Speech Pipelines and Prosodic Modeling
5. Voice Cloning and Speaker Adaptation
6. Singing and Music Generation

Chapter 5: Video Generation and Manipulation

1. Video Autoencoders and Temporal Modeling
2. Neural Rendering and 3D-Aware Generation
3. Face Swapping: AutoENCoders, StyleGAN-Based Methods, and SimSwap
4. Lip-Sync and Talking-Head Synthesis
5. Temporal Consistency Challenges
6. Video Compression Interaction with Generative Artifacts

Chapter 6: Digital Forensics and Signal Processing Fundamentals

1. Digital Evidence: Acquisition, Preservation, and Chain of Custody
2. Image Forensics Fundamentals: Error Level Analysis, Noise Patterns
3. Frequency-Domain Analysis: DFT, DCT, Wavelet Transforms
4. Statistical Analysis: Hypothesis Testing, Distribution Fitting, Anomaly Detection
5. Metadata Standards: EXIF, XMP, ID3, Forensic Metadata
6. File Format Analysis and Structural Artifacts

Chapter 7: Text Detection Part One: Statistical and Linguistic Signals

1. Perplexity and Likelihood-Based Detection
2. Burstiness, Entropy, and N-gram Statistics
3. Syntactic Complexity and Sentence Structure Patterns
4. Semantic Coherence and Topic Consistency
5. Stylometric Features and Authorship Attribution
6. Implementation: Statistical Text Detector

Chapter 8: Text Detection Part Two: Machine Learning and Transformer-Based Methods

1. Supervised Classifiers: SVMs, Random Forests on Text Features
2. Fine-Tuned Transformer Detectors: Architecture and Training
3. Zero-Shot and Few-Shot Detection with Prompt Engineering
4. Embedding-Based Detection and Clustering Approaches
5. Source Attribution: Which Model Generated This Text?
6. Implementation: Fine-Tuned Transformer Text Detector

Chapter 9: Image Detection Part One: Classical and Statistical Forensics

1. Error Level Analysis and Compression Mismatch Detection
2. Noise Residual Analysis and PRNU
3. Frequency-Domain Anomalies in Generated Images
4. Edge and Contour Artifacts
5. Illumination and Shadow Consistency Checks
6. Implementation: Error Level Analysis and Noise Residual Detector

Chapter 10: Image Detection Part Two: Deep Learning-Based Detection

1. CNN-Based Classifiers for GAN-Generated Image Detection
2. Frequency-Domain CNN Features
3. Patch-Based and Local Inconsistency Detection
4. Transformer-Based Image Detectors
5. Diffusion-Specific Detection: Noise Fingerprints and Sampling Artifacts
6. Implementation: CNN-Based Image Generator Detector

Chapter 11: Image Detection Part Three: Metadata, Provenance, and Watermarking

1. Metadata Analysis: EXIF, XMP, and File Signatures
2. Image Editing History Reconstruction
3. Invisible Watermarking: Spread-Spectrum and Steganographic Methods
4. Robust Watermarking for Diffusion Models
5. C2PA and Content Credentials Standard
6. Implementation: Metadata Forensics and Watermark Analysis Pipeline

Chapter 12: Video Detection

1. Frame-Level Detection: Applying Image Detectors to Video
2. Temporal Inconsistency Analysis: Flickering, Blending Artifacts
3. Facial Biometric Cues: Blinking, Eye Contact, Micro-Expressions
4. Audio-Video Sync and Lip-Sync Verification
5. Deepfake-Specific Artifacts: Boundary Regions, Skin Tone Inconsistencies
6. Implementation: Frame-Level Plus Temporal Video Detector

Chapter 13: Audio Detection

1. Spectral Analysis and Anomaly Detection in Synthetic Speech
2. Prosodic and Temporal Pattern Analysis
3. Voice Biometric Verification and Speaker Comparison
4. Neural Artifact Detection: Vocoder Fingerprints and Spectral Residuals
5. Zero-Shot Speaker Verification for Deepfake Detection
6. Implementation: Spectral Analysis Speech Deepfake Detector

Chapter 14: Multimodal Detection and Cross-Modal Consistency

1. Visual-Text Consistency: Does the Image Match the Caption?
2. Audio-Visual Consistency: Does the Speech Match the Video?
3. Multimodal Transformers for Joint Detection
4. Cross-Modal Retrieval and Embedding Consistency
5. Exploiting Generative Model Limitations in Multimodal Tasks
6. Implementation: Multimodal Consistency Checker

Chapter 15: Watermarking and Cryptographic Provenance

1. Visible vs. Invisible Watermarking: Trade-offs and Applications
2. Spread-Spectrum Watermarking for Images and Audio
3. Perceptual Hashing and Content-Based Authentication
4. Cryptographic Signatures and the C2PA Framework
5. Content Credentials: Implementation and Limitations
6. Blockchain-Based Provenance and Its Practical Constraints

Chapter 16: Source Attribution and Model Fingerprinting

1. Architectural Fingerprints in Outputs
2. Training Data Leakage and Memorization Artifacts
3. Statistical Signature Analysis for Source Identification
4. Embedding-Based Model Fingerprinting
5. Challenges: Model Variants, Fine-Tuning, and Ensemble Generators
6. Case Study: Distinguishing Between Major Image Generators

Chapter 17: Evaluation Methodology

1. Confusion Matrix Metrics: Accuracy, Precision, Recall, F1, ROC-AUC, PR-AUC
2. Cost-Sensitive Evaluation: False Positives vs. False Negatives
3. Calibration: Confidence Scores, Reliability Diagrams, Temperature Scaling
4. Robustness Evaluation Across Generators, Domains, and Transformations
5. Dataset Contamination, Benchmark Leakage, and Evaluation Integrity
6. Implementation: Complete Detector Evaluation Pipeline

Chapter 18: Adversarial Attacks and Robustness

1. Transformation-Based Attacks: Compression, Resizing, Denoising, Cropping
2. Text-Specific Attacks: Paraphrasing, Translation, Style Transfer
3. Adversarial Perturbations: Gradient-Based and Optimization-Based Attacks
4. Prompt Engineering as an Evasion Technique
5. Defenses: Ensemble Methods, Frequency-Domain Robustness, Watermarking
6. Implementation: Adversarial Robustness Testing Framework

Chapter 19: Limitations and Fundamental Constraints

1. Distribution Shift and Generalization Failure
2. The Fundamental Impossibility of Universal Detection
3. Degradation from Post-Processing and Editing
4. Human vs. AI Boundary Blurring with Advanced Models
5. Legal, Ethical, and Societal Constraints on Detection Deployment
6. What We Cannot Know: Fundamental Epistemic Limits

Chapter 20: Real-World Forensic Workflows

1. Evidence Acquisition and Preservation Procedures
2. Chain of Custody and Documentation
3. Triage and Automated Screening
4. Manual Forensic Review and Multi-Method Analysis
5. Confidence Reporting and Expert Testimony Considerations
6. Case Study: Investigating a Suspected AI-Generated Video

Chapter 21: Tools, Datasets, Standards, and Ecosystem

1. Open-Source Detection Tools and Libraries
2. Major Datasets: Text, Image, Video, and Audio Benchmarks
3. Standards Organizations: IEEE, ISO, NIST Efforts
4. Commercial Detection Systems and Their Claims
5. Research Communities and Preprint Culture
6. Reproducibility Crisis in Detection Research

Chapter 22: Future Directions and Responsible Practice

1. Emerging Generative Techniques and Their Detection Challenges
2. Active Provenance: Watermarks, Credentials, and Policy
3. Human-AI Collaboration in Forensic Analysis
4. Responsible Disclosure and Red-Teaming
5. Policy Implications and Governance Frameworks
6. Concluding Synthesis: Detection as Practice, Not Product

Conclusion

References