A Clear Chain of Evidence
How VeriTrace Works
VeriTrace uses a multi-stage verification pipeline to detect manipulated media, extract verifiable claims, and corroborate evidence against primary sources.
Detect
Uploaded media is hashed using SHA-256 for instant cache verification. New media is analyzed by Gemini multimodal models to detect synthetic generation indicators:
- Facial geometry and boundary anomalies
- Lighting & shadow direction inconsistencies
- Audio spectral cuts and voice synthesis cues
- Temporal frame jitter and lip-sync mismatch
Explain
Rather than a black-box percentage, VeriTrace breaks down the reasoning into verifiable visual and forensic indicators:
- Severity rankings: High, Medium, and Low
- Plain-language forensic explanations
- Region-specific markers highlighted on media
- Confidence score calculated from composite signals
Verify & Trace
Visual inspection is paired with factual grounding. The pipeline extracts real-world claims and runs Google Search grounding:
- Claim extraction from captions & scene context
- Google Search grounding cross-referencing
- Authoritative source citation (Reuters, AP, official archives)
- Context verdict: Corroborated or Contradicted
Core Principles & Transparency
Evidence, Not Proof
AI detection is not definitive proof. Our platform empowers researchers, journalists, and citizens by surfacing indicators so humans can make informed judgments.
Privacy & Efficiency
Media is hashed with SHA-256 for instant deduplication. Identical files resolve in under 50ms without redundant AI inference.
