AI & Deepfake Detection System

LunaNet

Revealing the Unseen

CNN Architecture FFT Frequency Analysis Deepfake Detection Deep Learning
· ·
Transmit your image for analysis
Drag & drop · or select a file · JPG · PNG · WEBP · max 10 MB
preview
Reading the stars of your image…
Image Ingestion
CNN Extraction
FFT Spectrum
Artifact Scan
Classification
Initialising neural pathways…
Detection Result
AI Generated
Deepfake
Authentic
Signal Analysis Visualisation
CNN Activation Map
FFT Frequency Spectrum
✦  Why It Matters  ✦
The Hidden Threat

Synthetic Images Are
Reshaping Reality

AI-generated and deepfake images now flood social media, news ecosystems, and personal communications at unprecedented scale. As generative models grow more capable, the perceptual boundary between real and fabricated has effectively collapsed. LunaNet is trained to see what the human eye cannot.

01
Non-Consensual Imagery
Deepfake technology is weaponised to fabricate intimate images of real individuals — predominantly women — causing severe psychological, reputational, and legal harm at scale.
02
Disinformation Campaigns
Synthetic images of political figures and public events are manufactured and distributed at scale, eroding trust and influencing elections and policy decisions.
03
Financial Fraud
AI-generated faces are deployed in identity fraud, fake KYC verification, and impersonation scams — resulting in billions in losses for individuals and institutions.
04
Judicial Evidence Risks
Fabricated images submitted as legal evidence represent a fundamental threat to justice systems when undetectable by conventional forensic methods.
05
Erosion of Trust
When authentic images become indistinguishable from fakes, public trust in visual media collapses — creating an environment where nothing can be verified.
06
Child Safety Threats
Generative AI is misused to produce synthetic harmful imagery, creating an unprecedented crisis for law enforcement and child protection agencies worldwide.
✦  Theory  ✦
Understanding the Source

How Synthetic Images
Are Generated

Modern AI-generated images emerge from distinct generative paradigms — each leaving unique statistical fingerprints in pixel data, texture gradients, and frequency spectra that LunaNet is trained to identify.

01
Generative Adversarial Networks (GANs)
A generator and discriminator network compete in a minimax game. At convergence the generator produces photorealistic outputs — but GAN upsampling operations leave periodic spectral artifacts in the FFT domain, particularly at mid-to-high frequencies. These are a primary detection signal for LunaNet.
02
Diffusion Models (Stable Diffusion, DALL·E)
Diffusion models learn to denoise images from Gaussian noise toward a target distribution. Despite perceptual quality, they produce characteristic pixel-level texture inconsistencies — unnatural skin smoothness, hair regularity, and specular anomalies — detectable through CNN spatial analysis.
03
Variational Autoencoders (VAEs)
VAEs encode images into a compressed latent space and decode sampled vectors into new outputs. They produce slightly blurrier images than GANs, with distinctive low-frequency spectral signatures. Many modern architectures combine VAEs with diffusion in a latent diffusion framework.
04
Face-Swap Deepfakes
Deepfake pipelines transplant facial identity onto a donor frame using encoder-decoder networks. They generate blending artifacts at face boundaries, temporal inconsistencies in blink and lip-sync, and abnormal colour distribution in skin regions — all detectable through LunaNet's spatial CNN stage.
✦  Technology  ✦
Detection Engine

LunaNet's Dual-Domain
Analysis Pipeline

CNN
Convolutional Neural Network
LunaNet's CNN backbone extracts hierarchical spatial features across multiple resolution scales. Early layers capture low-level texture anomalies — unnatural smoothness, repetitive micro-patterns, and noise inconsistencies. Deeper layers encode semantic irregularities: implausible facial geometry, abnormal eye reflections, and boundary artifacts at face-to-background transitions. The architecture draws from EfficientNet-style compound scaling for optimal accuracy and inference speed.
FFT
Fast Fourier Transform Analysis
The FFT stage transforms image pixel data into the frequency domain, exposing spectral artifacts imperceptible to the human eye but mathematically distinct in synthetic content. GAN generators produce characteristic frequency peaks from upsampling operations — the "checkerboard artifact" phenomenon. LunaNet computes the 2D magnitude spectrum and fuses radially-averaged frequency profiles with CNN spatial features in a joint classification head for superior accuracy.
Detection Pipeline
Input Image Preprocessing CNN Feature Maps FFT Spectrum Feature Fusion Classification Head Verdict + Confidence