vankey/DocShield-9B
DocShield-9B by vankey is a 9 billion parameter forensic-grade vision-language model, fine-tuned from Qwen3.5-VL-9B with a 32768 token context length. It specializes in document and image forgery analysis, providing structured reports with localized tampered regions, per-region reasoning, and a fraud-risk score. This model excels at detecting visual tampering traces, logical inconsistencies, and semantic alterations in documents.
Loading preview...
DocShield-9B: Forensic Document Forgery Analysis
DocShield-9B is a specialized 9 billion parameter vision-language model developed by vankey, designed for forensic-grade document and image forgery analysis. Fine-tuned from Qwen3.5-VL-9B, it leverages supervised training on the RealText forensic document datasets (RealText-V1, RealText-V2) to identify and report on document alterations. The model supports Qwen3.5's thinking mode for enhanced reasoning.
Key Capabilities
- Visual Forgery Trace Analysis: Detects subtle visual inconsistencies such as crude redactions, font/anti-aliasing issues, edge halos, copy-paste artifacts, and compression mismatches.
- Logical & Fact-Checking: Identifies contradictions in data like price/quantity, date conflicts, and bulk-discount logic violations.
- Semantic Alteration Detection: Pinpoints subtle changes in specifications (e.g., color, material) that might bypass basic visual checks.
- Localization: Provides bounding-box coordinates for each tampered region, accompanied by specific reasoning for each anomaly.
- Structured Reporting: Generates a comprehensive report including an overall conclusion (FORGED/authentic) and a fraud-risk score.
When to Use DocShield-9B
This model is ideal for applications requiring automated, detailed analysis of document authenticity. It's particularly suited for:
- Financial institutions verifying loan applications or invoices.
- Legal and compliance departments assessing document integrity.
- Insurance companies detecting fraudulent claims.
- Any scenario where robust, forensic-level detection of document tampering is critical. Its ability to provide localized evidence and a fraud-risk score makes it a powerful tool for automated fraud detection and document verification workflows.