Image Engine

The Image Engine verifies claims about images using deterministic methods first, with VLM fallback only for complex semantic claims.

Features

• Metadata Extraction - Dimensions, format (PNG, JPEG, GIF, WebP)
• Size Verification - 100% accurate dimension claims
• Claim Classification - Route to appropriate verifier
• Multi-VLM Consensus - Cross-validate semantic claims

Usage

from qwed_sdk import QWEDClient

client = QWEDClient(api_key="qwed_...")

# Read image
with open("chart.png", "rb") as f:
    image_bytes = f.read()

# Verify claim
result = client.verify_image(
    image=image_bytes,
    claim="The image is 800x600 pixels"
)

print(result.verdict)      # "SUPPORTED" or "REFUTED"
print(result.confidence)   # 1.0 (deterministic!)
print(result.methods_used) # ["metadata_extraction", "size_verification"]

Claim Types

Claim Type	Method	Deterministic?
Size/Dimensions	Metadata parsing	✅ 100%
Format	Header detection	✅ 100%
Color	Pixel sampling	⚠️ Partial
Text	OCR required	❌ VLM
Semantic	Understanding	❌ VLM

Multi-VLM Consensus

For semantic claims, use multiple VLMs:

result = client.verify_image_consensus(
    image=image_bytes,
    claim="The person is smiling",
    min_agreement=2  # At least 2 VLMs must agree
)

print(result.agreement_count)  # 3/3
print(result.vlm_results)      # Individual VLM responses