International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 13 Issue: 05 | May 2026
www.irjet.net
p-ISSN: 2395-0072
DETECTION OF AI-GENERATED IMAGES VS REAL IMAGES USING EFFICIENTNET-B0 Seerapu Yamini1, Vantakula Lakshmi2 1 MCA Student, Gayatri Vidya Parishad College of Engineering(A), Visakhapatnam – 530048, Andhra Pradesh,
India 2 Assistant Professor, Department of Computer Applications, Gayatri Vidya Parishad College of Engineering(A),
Visakhapatnam – 530048, Andhra Pradesh, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Artificial intelligence-generated images are
I. INTRODUCTION
rapidly increasing due to advancements in deep learning models and image synthesis technologies. Large volumes of
Artificial intelligence plays a vital role in transforming
synthetic images are produced every year using techniques
digital content creation, image processing, and media
such as Generative Adversarial Networks (GANs) and
generation
diffusion models. However, most of these images are highly
technologies such as Generative Adversarial Networks
realistic and difficult for users to distinguish from real
(GANs), diffusion models, and deep learning architectures
photographs without advanced analytical methods. This
are capable of producing highly realistic synthetic images
research presents the development of an AI-generated image
that closely resemble real-world photographs. These
detection system using EfficientNet-B0 and BLIP to identify
advanced technologies are widely used in fields including
whether an image is authentic or synthetically created by
entertainment, social media, advertising, virtual reality, and
artificial intelligence. The proposed system integrates
digital design.
EfficientNet-B0 with transfer learning to analyze hidden
Organizations and online platforms continuously generate
visual patterns, structural inconsistencies, and image
and share large volumes of digital images through social
artifacts for accurate classification. The system also
media applications, websites, and AIbased content creation
incorporates the BLIP (Bootstrapping Language-Image
systems. These images include both authentic photographs
Pretraining) model to generate descriptive captions for
and AIgenerated synthetic content created using advanced
uploaded images, improving interpretability and user
image generation techniques. Although such technologies
understanding. The system converts image data into
provide
classification results and textual descriptions through image
automation, they also create serious challenges related to
preprocessing, feature extraction, and caption generation. By
misinformation, deepfakes, image manipulation, and digital
transforming complex deep learning processes into a simple
authenticity verification.
and accessible solution, the proposed system simplifies image
Traditional image verification methods mainly rely on
authenticity verification and helps reduce the spread of
manual inspection and basic image analysis techniques.
misleading AI-generated content across digital media
This makes it difficult for users to accurately distinguish
platforms.
between real and AI-generated images, especially when
across
significant
various
domains.
advantages
in
Modern
creativity
AI
and
synthetic images contain highly realistic textures, lighting,
Key Words: AI-Generated Images, EfficientNet-B0,
and visual patterns. With the increasing availability of AI-
Deep Learning, Image Classification, BLIP, Image Authenticity Detection
© 2026, IRJET
|
Impact Factor value: 8.315
generated content, there is a growing need for intelligent
|
ISO 9001:2008 Certified Journal
|
Page 1004