Synthetic Photography Detection: A Visual Guidance for Identifying Synthetic Images Created by AI

TL;DR

This paper addresses the risk of photorealistic AI-generated synthetic photographs and proposes an artifact-based framework to identify synthetic origin. It introduces a taxonomy of artifact types derived from diffusion-model flaws and demonstrates spotting strategies across People, Indoor, Outdoor, and Objects, with domain-specific cues. It discusses challenges such as subtle artifacts and cognitive biases, and suggests practical applications in education, forensics, media literacy, and policy, along with future research directions in detection technology and bias mitigation. The work offers actionable guidance for educators, analysts, and researchers to improve resilience against misinformation and deception involving synthetic imagery.

Abstract

Artificial Intelligence (AI) tools have become incredibly powerful in generating synthetic images. Of particular concern are generated images that resemble photographs as they aspire to represent real world events. Synthetic photographs may be used maliciously by a broad range of threat actors, from scammers to nation-state actors, to deceive, defraud, and mislead people. Mitigating this threat usually involves answering a basic analytic question: Is the photograph real or synthetic? To address this, we have examined the capabilities of recent generative diffusion models and have focused on their flaws: visible artifacts in generated images which reveal their synthetic origin to the trained eye. We categorize these artifacts, provide examples, discuss the challenges in detecting them, suggest practical applications of our work, and outline future research directions.

Figures (25)

• Figure 1: Synthetic photograph with annotated artifacts: The hands of the two children around the neck seem merged. The baby's foot is missing toes, the boy who walks has indistinct feet, and there is an extra leg / or leg growing from the right shoulder of the man. This example demonstrates a typical weakness of generative models: The accurate depiction of human anatomy and details. Image Source: https://donya-e-eqtesad.com/%D8%A8%D8%AE%D8%B4-%D8%B3%D8%A7%DB%8C%D8%AA-%D8%AE%D9%88%D8%A7%D9%86-62/4014731-%D8%A7%DB%8C%D9%86-%D8%B9%DA%A9%D8%B3-%D8%AF%D9%86%DB%8C%D8%A7-%D8%B1%D8%A7-%D8%AA%DA%A9%D8%A7%D9%86-%D8%AF%D8%A7%D8%AF-%D8%AA%D8%B5%D9%88%DB%8C%D8%B1
• Figure 2: Synthetic image of each artifact category. \ref{['fig:artifact_overview:distortion']} and \ref{['fig:artifact_overview:text']} are from the Synthbuster dataset bammey_synthbuster_2023, created with Midjourney-V5.
• Figure 3: Artifact taxonomy. Distortion, shape, and depth examples from Synthbuster bammey_synthbuster_2023, generated with Midjourney v5.
• Figure 4: Synthetic portrait photography of a woman with small irregularities in the left eye, hair, teeth, and ear form.
• Figure 5: Eye artifacts can best revealed by zooming in. The picture at the right shows a real photograph as a comparison (https://unsplash.com/photos/a-woman-standing-in-front-of-a-sculpture-O3D-teBz0Bg).