ASAP: Visual Analytics for Identifying and Analyzing Image Patterns in AI-generated Images

Authors: Jinbin Huang, Yuki Ueno, Chen Chen, Aditi Mishra, Bum Chul Kwon, Zhicheng Liu, Chris Bryan

Published: 2026-09-23 05:20:13+00:00

Journal Ref: Information Visualization (2026)

AI Summary

ASAP is an interactive visualization system that helps users analyze and summarize deceptive patterns in AI-generated images. It introduces a novel CLIP-adapted image encoder that generates interpretable representations, enabling the extraction of influential pixel regions for identifying key deceptive features through influence measurement techniques. The system integrates these techniques into a visual analytics dashboard for quantifying and analyzing authenticity-indicative patterns in image collections, supporting comparative analysis of various generative models.

Abstract

Generative image models can produce highly realistic images, raising concerns about potential misuse in creating deceptive content. Current deepfake approaches face several challenges, including limited generalizability, lack of interpretability, and poor actionability. To help address these, we present ASAP, an interactive visualization system designed to empower users in the analysis and summarization of deceptive patterns in AI-generated images. ASAP introduces a novel CLIP-adapted image encoder that generates interpretable representations, enabling the extraction of influential pixel regions via calculated masks. This approach facilitates the identification of key deceptive features through influence measurement techniques. These backend techniques are integrated into a visual analytics dashboard that allows users to quantify and analyze authenticity-indicative patterns in image collections containing both authentic and AI-generated images. This approach also supports the comparative analysis of various generative models, including GANs and diffusion models. We demonstrate ASAP's efficacy through a user study and two application scenarios using established fake image detection benchmarks, showcasing its ability to effectively extract and quantify deceptive patterns.

Key findings

ASAP effectively extracts and quantifies deceptive patterns in AI-generated images, demonstrating strong generalizability and interpretability. A user study and application scenarios confirmed its ability to support detailed pattern analysis and comparison across different generative models. The system helps users understand why images are classified as real or fake by revealing influential pixel regions and their contributions to predictions.

Approach

ASAP addresses challenges in deepfake detection by using a human-in-the-loop visual analytics approach. It trains a binary classifier on a CLIP-adapted image encoder to distinguish real from fake images, learning interpretable representations by filtering out textual information and distilling authenticity-indicative dimensions. Influence measurement techniques identify critical pixel regions, which are then presented in an interactive visual analytics dashboard to analyze and quantify deceptive patterns.

Datasets

proGAN dataset, DetectingSyntheticImage dataset (specifically LDM human face images from Flickr-Faces-HQ (FFHQ))

Model(s)

CLIP:ViT-B/32 (adapted with a 'forget-to-spell' projection and a distiller layer)

Author countries

← Previous