This paper introduces LiDAR Diffusion Models (LiDMs), a novel approach for generating realistic LiDAR scenes. LiDMs are designed to address the challenges of generating high-quality, realistic LiDAR data using diffusion models, which have shown great success in image synthesis. The key innovations of LiDMs include curve-wise compression to preserve LiDAR patterns, point-wise coordinate supervision to learn scene geometry, and patch-wise encoding to capture the full context of 3D objects. These components enable LiDMs to generate LiDAR-realistic scenes with high efficiency and accuracy, achieving state-of-the-art results in unconditional LiDAR generation. Additionally, LiDMs support conditional generation from various input modalities, such as semantic maps, camera views, and text prompts, enhancing their versatility and applicability in autonomous driving and robotics. The method is evaluated on both 32-beam and 64-beam LiDAR data, demonstrating significant improvements over existing methods in terms of performance and efficiency. The results show that LiDMs can generate high-quality LiDAR scenes with a 107x speedup compared to point-based diffusion models, making them a promising solution for realistic LiDAR scene generation.This paper introduces LiDAR Diffusion Models (LiDMs), a novel approach for generating realistic LiDAR scenes. LiDMs are designed to address the challenges of generating high-quality, realistic LiDAR data using diffusion models, which have shown great success in image synthesis. The key innovations of LiDMs include curve-wise compression to preserve LiDAR patterns, point-wise coordinate supervision to learn scene geometry, and patch-wise encoding to capture the full context of 3D objects. These components enable LiDMs to generate LiDAR-realistic scenes with high efficiency and accuracy, achieving state-of-the-art results in unconditional LiDAR generation. Additionally, LiDMs support conditional generation from various input modalities, such as semantic maps, camera views, and text prompts, enhancing their versatility and applicability in autonomous driving and robotics. The method is evaluated on both 32-beam and 64-beam LiDAR data, demonstrating significant improvements over existing methods in terms of performance and efficiency. The results show that LiDMs can generate high-quality LiDAR scenes with a 107x speedup compared to point-based diffusion models, making them a promising solution for realistic LiDAR scene generation.