Aiming at the non-Gaussian noise problem, which is common in image processing, a new method of generalized diffusion model for non-Gaussian noise was proposed in this paper. The classical Diffusion Model is trained and reasoned only on Gaussian noise, while Cold Diffusion can use arbitrary noise to process images. We use images with Gaussian and non-Gaussian noise (e.g., uniform noise, salt-and-pepper noise) to train models and use the models to reduce the noise. The results show that the generalized diffusion model can restore more features and details when dealing with non-Gaussian noise. It is more robust and adaptable than the Denoising Diffusion Probabilistic Models. Noise could be treated as a generalized transform. Besides Gaussian noise, the model could adapt to any noise type instead of depending on a specific type of noise. As an emerging generative model, the generalized diffusion model offers new ideas and methods to solve complex inverse problems in the real world.