Eviscene assesses image truthfulness with evidence based AI
EviScene: Beyond "Is It AI?" for Agentic Image Factuality Assessment
Computer Vision and Pattern Recognition
Summary
Identifying whether an image is generated by AI doesn’t tell us if the scene shown is actually true. The authors introduce a way to check if the situation in an image is real or not, using both the picture and external information. Their system, called EviScene, asks smart questions, searches for evidence, and explains its decisions. It was tested on over 10,000 images and showed good accuracy in deciding if an image’s story is supported by facts and evidence.
What this means in practice
- •For media verification teams: Assess the factual truth of images in news or social media posts to help detect misinformation with an explanation grounded in evidence.
- •For social media platform moderators: Automatically verify if content images are factually accurate, improving content moderation tools to reduce false or misleading posts.
Authors
ZiAn Wang, Ruiqi Liu, Zhengbo Zhang, Yi Han, Jialiang Shen, Ruoxin Chen, Manni Cui, Ximo Zhu, Jin Jiang, Mingzhe Liu, Linhao Wang, Wenhao Wang, Shu Wu
Abstract
Determining whether an image is AI-generated does not establish whether the situation it depicts is factually supported. We formulate Image Factuality Assessment: given an image and a generic verification instruction, a system returns a supported or refuted verdict with an evidence-grounded explanation. Without a supplied textual claim, the system must progressively refine factual questions during investigation and align external evidence with the specific situation depicted. We introduce EviLens, an expert-reviewed benchmark of 10,174 supported and refuted images from controlled construction and web collection, covering twelve factual error types and evaluating both verdict accuracy and evidence sufficiency. We develop EviScene, a multimodal ReAct agent that interleaves visual inspection, external retrieval, and evidence analysis, using new findings to refine its questions and revisit relevant image details. We train a compact policy with supervised fine-tuning on reviewed teacher trajectories, followed by privileged on-policy self-distillation applied to verified decisions from the policy's own investigations. On the 1,527-image test set, the trained Qwen3.5-9B agent achieves 79.05% balanced accuracy and a Strict Evidence Sufficiency Rate (SESR) of 52.26%, improving over the same-backbone base agent by 6.71 and 9.37 percentage points, respectively. Its SESR is the highest among the evaluated agents.Code, models, and data will be released at https://github.com/wangzian828/EVISCENE and https://huggingface.co/ZiAnwang/EviScene.