Five-category artifact taxonomy for evaluating AI-generated images

Design · All designs
Evidence · 2 claims (2 for) · 1 study (1 causal), q2 · 0 of 1 report an effect size · 2 claims rest on one study

Description

The article organizes visual inspection of AI-generated images around a taxonomy of five artifact categories drawn from a written guide on distinguishing AI-generated images from authentic photographs. The training "was structured around five categories of artifacts": anatomical implausibilities (hands, eyes, teeth, body structure, identity inconsistencies), stylistic artifacts (cinematization, plastic textures, hyper-real detail), functional implausibilities (dysfunctional objects, malformed text and logos), violations of physical constraints (inconsistent lighting, shadows, reflections, perspective), and sociocultural implausibilities (historically or culturally incongruent scenarios). Each category was taught with concise explanations paired with multiple visual examples, and the same taxonomy coded analysts' free-text reasoning in the qualitative analysis.

Design Implications

Context

Requirements

• Explanations paired with multiple visual examples per category, with visual callouts highlighting regions of interest
• Examples drawn from contemporary diffusion models so the taught cues match the artifacts current generators produce

Constraints

• The article cautions analysts against over-reliance on any single cue, since artifact visibility varies with scene complexity, pose, image resolution, and viewing time
• Real photographs can contain non-generative artifacts such as compression and focus inconsistencies, so the presence of an artifact does not by itself imply an image is AI-generated

Target Learners

• intelligence analysts
• professionals whose verification decisions carry operational weight, such as journalists and content moderators

Learning Goals

• distinguishing real photographs from AI-generated images
• articulating detection reasoning using specific artifact categories

Claims

• Training Shifts Analyst Detection Reasoning Toward Specific Artifact Vocabulary [+M]
• Brief Generative Ai Literacy Training Raises Analyst Image Discrimination Accuracy [+M]

Related Designs

Examples

-

Key Sources

• Negar Kamali, Candice Rockell Gerstner, Jessica Hullman, Matthew Groh. (2026). Generative AI Literacy Training Improves Intelligence Analysts' Discrimination of Real and AI-Generated Images. https://arxiv.org/abs/2606.28510

Page metadata

Field	Value
type	design
id	five-category-ai-image-artifact-taxonomy
title	Five-category artifact taxonomy for evaluating AI-generated images
description	The article organizes visual inspection of AI-generated images around a taxonomy of five artifact categories drawn from a written guide on distinguishing AI-generated images from authentic photographs.
status	review
generated	by process:settle-candidates on 2026-10-11
trust	unverified — no one has confirmed this page's content
sources	1 — negar-kamali-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.