SEED: Simple ViT and Evolving Harness for Explainable Text Forgery Detection

2026-06-19

AI Summary

This work addresses the growing threat posed by AI-generated text-image forgeries to the integrity of financial, legal, and identity records by introducing SEED, a novel forensic system that uniquely integrates detection and pixel-level localization within a single Vision Transformer (ViT). Leveraging a DINOv3 backbone fine-tuned with LoRA, SEED preserves rich pre-trained priors while enabling precise forgery identification. To enhance robustness, the system employs similarity-guided data augmentation to synthesize diverse forged samples. Furthermore, it incorporates a multimodal large language model (MLLM) with a proposal-evaluation iterative mechanism to produce structured, multilingual, and interpretable forensic reports. The approach achieved third place in the ACM MM 2026 GenText-Forensics Challenge, and both code and datasets have been publicly released.

Abstract

AI-assisted image editing threatens trust in financial, legal, and identity records. The GenText-Forensics Challenge at ACM MM 2026 addresses this by requiring structured forensic reports, in which integrating detection, pixel-level localization, and natural language explanation for multilingual text-centric forgery images. We present SEED, a modular system with three components. First, a similarity-guided pipeline augments training with diverse synthetic forgeries. Second, a single ViT, built on DINOv3 with LoRA adaptation, jointly performs detection and pixel-level localization while preserving pre-trained priors with minimal trainable parameters. Third, an evolving harness takes the detector's predictions and generates a complete forensic report via an MLLM, iteratively improved through a proposer-evaluator loop optimizing report quality. SEED ranked 3rd in the GenText-Forensics Challenge. Code and data are available at https://github.com/KahimWong/GenText-Forensics-3rd-Place.

Problem

• text forgery detection
• explainable AI
• pixel-level localization
• multilingual forensic report
• AI-generated image

Innovation

• Vision Transformer
• LoRA adaptation
• explainable forgery detection
• multimodal LLM
• synthetic data augmentation