SemCity is a 3D diffusion model designed for generating semantic scenes in real-world outdoor environments. The model uses a triplane representation to address the challenges of generating real-world outdoor scenes, which often contain more empty spaces due to sensor limitations. The triplane representation helps reduce unnecessary empty information by factorizing 3D data into three orthogonal 2D planes, making it more efficient for capturing relevant spatial details. The model is trained on real-world outdoor datasets such as SemanticKITTI and can be extended to various practical tasks like scene inpainting, scene outpainting, and semantic scene completion refinement. The triplane diffusion model generates new scenes by creating novel triplanes based on the efficient representation. The model also allows for seamless manipulation of triplanes to add, remove, or modify objects within a scene. The triplane diffusion model has shown meaningful generation results compared to existing work in real-world outdoor datasets. The model's ability to generate detailed scenes and handle various tasks makes it a promising approach for semantic scene generation in real-world outdoor environments.SemCity is a 3D diffusion model designed for generating semantic scenes in real-world outdoor environments. The model uses a triplane representation to address the challenges of generating real-world outdoor scenes, which often contain more empty spaces due to sensor limitations. The triplane representation helps reduce unnecessary empty information by factorizing 3D data into three orthogonal 2D planes, making it more efficient for capturing relevant spatial details. The model is trained on real-world outdoor datasets such as SemanticKITTI and can be extended to various practical tasks like scene inpainting, scene outpainting, and semantic scene completion refinement. The triplane diffusion model generates new scenes by creating novel triplanes based on the efficient representation. The model also allows for seamless manipulation of triplanes to add, remove, or modify objects within a scene. The triplane diffusion model has shown meaningful generation results compared to existing work in real-world outdoor datasets. The model's ability to generate detailed scenes and handle various tasks makes it a promising approach for semantic scene generation in real-world outdoor environments.