Introduction

Bijou API for AI-generated image detection

Overview

The Bijou API lets you detect whether images are AI-generated using an ensemble of state-of-the-art detection models. Our multi-model approach provides high accuracy by combining results from multiple specialized detectors.

Base URL

https://api.trybijou.com/v1

Detection Models

Bijou uses an ensemble of 5 detection models, each specialized for different AI generation techniques:

Model	Weight	Specialization
AEROBLADE	22%	VAE reconstruction artifacts
UnivFD	22%	CLIP-based feature analysis
NPR	18%	Neighboring pixel relationships
FIRE	22%	Frequency-based analysis
CNNSpot	16%	GAN fingerprint detection

Quick Example

Python

from bijou import BijouClient

client = BijouClient(api_key="bijou_your-key")
result = client.detect("path/to/image.jpg")

print(f"Prediction: {result.prediction}")  # "ai" or "real"
print(f"Confidence: {result.confidence:.1%}")

Response Format

All API responses follow this structure:

{
  "statusCode": 200,
  "body": {
    "data": {
      "prediction": "ai",
      "confidence": 0.87,
      "ensemble_score": 0.87,
      "verdict": "ai_generated",
      "models_succeeded": 5,
      "models_total": 5,
      "model_results": {
        "aeroblade": {"score": 0.92, "latency_ms": 145},
        "univfd": {"score": 0.88, "latency_ms": 132},
        "npr": {"score": 0.79, "latency_ms": 98},
        "fire": {"score": 0.91, "latency_ms": 156},
        "cnnspot": {"score": 0.82, "latency_ms": 112}
      },
      "detection_latency_ms": 523
    }
  }
}

Verdict Thresholds

Ensemble Score	Verdict
>= 0.7	ai_generated
0.4 - 0.7	uncertain
< 0.4	likely_real

Next Steps

Authentication

Set up your API key

Python SDK

Install and use the Python SDK

TypeScript SDK

Install and use the TypeScript SDK

Error Handling

Handle API errors gracefully