RSFAKE-1M: A Large-Scale Dataset for Detecting Diffusion-Generated Remote Sensing Forgeries

Detecting forged remote sensing images is critical for maintaining trust in environmental monitoring and national security. While diffusion models have enabled highly realistic forgeries, their impact on forensic systems remains underexplored due to the lack of adequate benchmarks. Existing datasets are typically small-scale and GAN-centric, failing to represent the complexity of modern generative models. To bridge this gap, we introduce RSFAKE-1M , a comprehensive million-scale benchmark containing 500K real and 500K forged images. It captures a wide spectrum of forgery traces by incorporating samples from ten distinct diffusion models under six diverse generation conditions (e.g., text-to-image, inpainting). To demonstrate the benchmark's utility and explore new detection paradigms, we conduct a systematic evaluation of existing detectors and introduce RSFAKE-bench , an exploratory MLLM-based baseline that utilizes token probabilities to derive a continuous detection score. Our extensive experiments on RSFAKE-1M reveal the limitations of current specialized detectors in cross-generator generalization, while the MLLM-based exploration indicates that RSFAKE-bench can additionally provide human-readable reasons describing geographic and physical inconsistencies for its judgments. We release RSFAKE-1M to serve as a foundational resource for developing and rigorously evaluating next-generation forensic solutions. The dataset is available at https://huggingface.co/datasets/TZHSW/RSFAKE/ .