This paper introduces ItDPDM, a Poisson-based diffusion model that directly models discrete data without relying on continuous embeddings or variational approximations, achieving improved likelihood estimation on music and image benchmarks.