9-DETECTOR ENSEMBLE · 89% ON MAY 2026 GENERATORS

Spot AI-generated images.
Before they fool anyone.

PixelPrism runs nine independent forensic detectors on every image — frequency analysis, three vision transformers, diffusion reconstruction, CLIP anomaly, sensor noise, EXIF/C2PA, and face deepfake — fused by a tree-based meta-classifier Beta retrained every time a major generator drops.

Try it free → View API Docs
89%
FRESH-AI ACCURACY (MAY 2026)
9
DETECTORS PER IMAGE
<350ms
END-TO-END LATENCY
$0.0005
PER IMAGE (PRO)
UNDER THE HOOD

Seven detectors. One verdict.

Each detector targets a different artifact class. The full ensemble is what makes us accurate across Midjourney, FLUX, DALL-E, Stable Diffusion, and face-swap deepfakes.

📡
FFT Frequency Analyzer
Detects GAN checkerboard patterns and diffusion sampling noise invisible to the eye.
CPU · ~5ms
🧠
ViT-Large Classifier
307M-parameter Vision Transformer fine-tuned on 1M+ real vs AI image pairs.
GPU · ~50ms
🔄
DIRE Reconstruction
Measures reconstruction error through a diffusion model — AI images live on the manifold.
GPU · ~200ms
🎯
CLIP Anomaly Score
Mahalanobis distance from real-image distribution. Generalizes to unseen generators.
GPU · ~30ms
🎚
SRM Noise Pattern
Spatial Rich Model filters extract camera sensor noise. Real photos have it, AI doesn't.
CPU · ~10ms
👤
Face Deepfake (Xception)
FaceForensics++ trained model. Activates when MediaPipe detects a face.
GPU · ~40ms
🪪
EXIF Metadata Checker
Inspects camera fingerprints and C2PA Content Credentials for provenance.
CPU · <1ms
⚡
Meta-Classifier Fusion
Logistic-regression layer learns the optimal weight for each detector — trained on labeled real vs AI data. This is what pushes accuracy above 95%.
WHO IT'S FOR

Built for anyone whose credibility depends on truth.

From newsroom verification to identity-platform trust & safety.

📰
Journalists & Newsrooms
Verify source images before publication. Detect AI-doctored evidence in seconds, not hours. Public-interest pricing available.
🛡️
Trust & Safety Teams
Flag AI-generated profile pictures, romance-scam imagery, and synthetic content at scale. Webhook + REST API for any moderation pipeline.
🏛️
KYC & Compliance
Catch deepfake selfies in identity verification. On-premise deployment for financial institutions, healthcare, and government — your data never leaves.
HOW IT WORKS

From upload to verdict in under 350ms.

Three steps, one API call.

STEP 01
Send the image
POST any image (JPEG, PNG, WebP, HEIC) to /detect via REST or upload it through the web UI.
STEP 02
Run the ensemble
All 7 detectors execute in parallel on the GPU. Their scores are fused by a meta-classifier trained on labeled real vs AI data.
STEP 03
Get a forensic report
JSON response with verdict, confidence, per-detector breakdown, and triggered signals. Webhook delivery available on Pro+.

Detect your first image in 2 minutes.

50 free detections every month. No card required. No setup friction.

→ Start Free See Pricing