This repository provides the official implementation of score-based generative modeling through stochastic differential equations, a unified framework that generalizes diffusion models. The method transforms data to noise via continuous-time stochastic processes and reverses the process for generation using learned score functions. It supports unconditional generation, class-conditional generation, inpainting, colorization, and latent code manipulation, achieving FID of 2.20 on CIFAR-10 and high-fidelity 1024px CelebA-HQ generation.