aiqbench analysis
This study demonstrates that frozen CLIP embeddings can effectively detect AI-generated images with 95% accuracy using lightweight classifiers, bypassing the need for language reasoning. However, the research highlights persistent vulnerabilities in classifying specific artistic styles like oil paintings and wide-angle photography, indicating gaps in current detection robustness. These findings are crucial for developers building content verification systems who must account for style-dependent failure modes in vision-language models.
CLIP Embeddings for AI-Generated Image Detection: A Few-Shot Study with Lightweight Classifier
arXiv:2505.10664v2 Announce Type: replace-cross Abstract: Verifying the authenticity of AI-generated images presents a growing challenge on social media platforms these days. While vision-language models (VLMs) like CLIP outdo in multimodal representation, their capacity for AI-generated image classification is underexplored due to the absence of such labels during the pre-training process. This work investigates whether CLIP embeddings inherently contain information indicative of AI generation. A proposed pipeline extracts visual embeddings using a frozen CLIP model, feeds its embeddings to lightweight networks, and fine-tunes only the final classifier. Experiments on the public CIFAKE benchmark show the performance reaches 95% accuracy without language reasoning. Few-shot adaptation to curated custom with 20% of the data results in performance to 85%. A closed-source baseline (Gemini-2.0) has the best zero-shot accuracy yet fails on specific styles. Notably, some specific image types, such as wide-angle photographs and oil paintings, pose significant challenges to classification. These results indicate previously unexplored difficulties in classifying certain types of AI-generated images, revealing new and more specific questions in this domain that are worth further investigation.
Abstract:Verifying the authenticity of AI-generated images presents a growing challenge on social media platforms these days. While vision-language models (VLMs) like CLIP outdo in multimodal representation, their capacity for AI-generated image classification is underexplored due to the absence of such labels during the pre-training process. This work investigates whether CLIP embeddings inherently contain information indicative of AI generation. A proposed pipeline extracts visual embeddings using a frozen CLIP model, feeds its embeddings to lightweight networks, and fine-tunes only the final classifier. Experiments on the public CIFAKE benchmark show the performance reaches 95% accuracy without language reasoning. Few-shot adaptation to curated custom with 20% of the data results in performance to 85%. A closed-source baseline (Gemini-2.0) has the best zero-shot accuracy yet fails on specific styles. Notably, some specific image types, such as wide-angle photographs and oil paintings, pose significant challenges to classification. These results indicate previously unexplored difficulties in classifying certain types of AI-generated images, revealing new and more specific questions in this domain that are worth further investigation.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2505.10664 [cs.CV]
(or arXiv:2505.10664v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2505.10664 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 267 words
Related coverage
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026
Detect Before You Leap: Mirage Detection in Vision-Language Models
arXiv · Jun 16, 2026
[AINews] not much happened today
Latent Space · Oct 3, 2026
How to Measure Video Similarity: 6 Techniques I Tested (and the One I Shipped)
Analytics Vidhya · Jul 13, 2026