Plenoxels is a method for photorealistic view synthesis that represents scenes as a sparse 3D grid with spherical harmonics. Unlike Neural Radiance Fields (NeRF), Plenoxels do not use neural networks, allowing for significantly faster optimization while maintaining high visual quality. The method uses a sparse voxel grid where each voxel stores opacity and spherical harmonic coefficients, enabling continuous representation of the plenoptic function. Plenoxels can be optimized two orders of magnitude faster than NeRF, with training times reduced from over a day to just minutes on a single GPU. The method supports both bounded and unbounded scenes, including 360-degree views, and can be converted into a PlenOctree for faster rendering. Plenoxels use trilinear interpolation for efficient rendering and optimization, and incorporate total variation regularization to improve quality. The method achieves state-of-the-art results without neural networks, demonstrating that photorealistic volumetric reconstruction can be achieved using standard inverse problem tools. Plenoxels outperform prior methods in speed and quality, with results shown on synthetic and real scenes. The method is implemented in CUDA and provides a fast, efficient solution for 3D reconstruction.Plenoxels is a method for photorealistic view synthesis that represents scenes as a sparse 3D grid with spherical harmonics. Unlike Neural Radiance Fields (NeRF), Plenoxels do not use neural networks, allowing for significantly faster optimization while maintaining high visual quality. The method uses a sparse voxel grid where each voxel stores opacity and spherical harmonic coefficients, enabling continuous representation of the plenoptic function. Plenoxels can be optimized two orders of magnitude faster than NeRF, with training times reduced from over a day to just minutes on a single GPU. The method supports both bounded and unbounded scenes, including 360-degree views, and can be converted into a PlenOctree for faster rendering. Plenoxels use trilinear interpolation for efficient rendering and optimization, and incorporate total variation regularization to improve quality. The method achieves state-of-the-art results without neural networks, demonstrating that photorealistic volumetric reconstruction can be achieved using standard inverse problem tools. Plenoxels outperform prior methods in speed and quality, with results shown on synthetic and real scenes. The method is implemented in CUDA and provides a fast, efficient solution for 3D reconstruction.