JiT is a pixel-space diffusion model that denoises images directly in pixel space rather than using latent compression. This PyTorch re-implementation reproduces the research paper “Back to Basics: Let Denoising Generative Models Denoise” by Li et al., originally implemented in JAX/TPU. The model supports training on ImageNet at 256x256 resolution with transformer architectures such as JiT-B/16.