EviScene: Beyond "Is It AI?" for Agentic Image Factuality Assessment

Key points

• Determining whether an image is AI-generated does not establish whether the situation it depicts is factually supported.1
• 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.1
• 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.1
• 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.1

Sources (2)

• [1]EviScene: Beyond "Is It AI?" for Agentic Image Factuality Assessment
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 10, 05:52 AM

  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.

  Determining whether an image is AI-generated does not establish whether the situation it depicts is factually supported.

• [2]EviScene: Beyond "Is It AI?" for Agentic Image Factuality Assessment
  Hugging Face Daily Papers · Oct 10, 12:00 AM · same content

Extractive summary: sentences quoted from the sources.