New GenSyn10 dataset benchmarks AI image detection across diverse generators

Researchers have introduced GenSyn10, a new dataset designed to benchmark the detection of AI-generated images. The dataset comprises 60,000 images aligned with CIFAR-10, created using three distinct generative models: FLUX.2-dev, HunyuanImage-3.0, and Qwen Image 2512. GenSyn10 aims to improve the out-of-distribution generalization of AI-generated image detectors, as current models often struggle with generators they haven't been trained on. While models fine-tuned on GenSyn10 achieve high accuracy on seen generators, their performance drops significantly when encountering images from unseen generators, highlighting the ongoing challenge in robust AI-generated image detection. AI

IMPACT This dataset will help researchers develop more robust AI image detection models capable of identifying outputs from a wider range of generative architectures.

RANK_REASON The item is a research paper introducing a new dataset for benchmarking AI image detection. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• CIFAR-10
• FLUX.2-dev
• GenSyn10
• Md Faraz Kabir Khan
• MoE Transformer
• Multimodal Diffusion Transformer
• Qwen Image 2512
• Rectified Flow transformer

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →