Model data through forward noising and learned reverse denoising processes. Implement a compact diffusion model, use conditional guidance, and compare sampling quality, diversity, and speed.
