98.8% AUC on 83K+ test images

Detect Deepfakes
with AI Precision

Upload any image and our dual-branch AI analyses spatial artifacts and frequency patterns to determine if a face has been digitally manipulated. Transparent results with Grad-CAM++ heatmaps.

Start DetectionHow It Works
EfficientNetV2-SFFT AnalysisGrad-CAM++3 Datasets21.6M params

Built for Real-World Performance

Trained on FaceForensics++, Celeb-DF v2, and Kaggle datasets with progressive unfreezing.

98.88%

AUC-ROC

Test set discrimination

96.85%

Accuracy

On 83K test images

566K+

Training Data

Face images analysed

21.6M

Parameters

Dual-branch model

Simple 3-Step Process

How It Works

01

Upload Image

Drag & drop or browse for any photo. Our system automatically detects and aligns the face in the image.

02

AI Analysis

The dual-branch model analyses spatial features via EfficientNetV2-S and frequency patterns via FFT spectrum — then fuses them with SE Attention.

03

Explainable Results

Get a clear REAL/FAKE verdict, confidence score, and Grad-CAM++ heatmap showing exactly which facial regions the model focused on.

Dual-Branch Architecture

Unlike single-network detectors, DeepShield uses two complementary branches that capture different manipulation artifacts — making it harder for sophisticated deepfakes to evade detection.

1Spatial Branch — EfficientNetV2-S extracts visual artifacts (blending, warping)
2Frequency Branch — FFT CNN detects spectral anomalies invisible to the eye
3SE Fusion — Squeeze-Excitation attention combines both for robust prediction
┌─────────────────────┐
│    Input Image      │
│   (224 × 224 × 3)  │
└──────────┬──────────┘
           │
     ┌─────┴─────┐
     ▼           ▼
┌─────────┐ ┌─────────┐
│ Spatial │ │   FFT   │
│ Branch  │ │ Branch  │
│ EffV2-S │ │ 4L CNN  │
│ (1280-d)│ │ (256-d) │
└────┬────┘ └────┬────┘
     │           │
     └─────┬─────┘
           ▼
    ┌─────────────┐
    │  SE Fusion  │
    │  (1536-d)   │
    └──────┬──────┘
           ▼
    ┌─────────────┐
    │ Classifier  │
    │ → P(fake)   │
    └─────────────┘

Ready to verify an image?

Upload any photo and get an instant deepfake analysis with explainable AI heatmaps. Free, fast, transparent.

Start Detection

Built with

PyTorchEfficientNetV2-SFastAPINext.jsRetinaFaceGrad-CAM++Tailwind CSS