📄 中文摘要
扩散模型通过逐步去除噪声来生成图像。首先，向清晰的照片中逐渐添加静态噪声，直到照片变成纯噪声。接着，学习如何逆转这一过程，逐步去除噪声，最终呈现出逼真的图像。这一过程可以看作是一种魔术，开始时是随机的静态噪声，经过清理，最终展现出清晰的图像。该模型的训练过程涉及从清晰的照片开始，逐步添加噪声，直到达到纯噪声的状态，从而学习每一步的变化。
🌫️ 扩散模型简单解释
出处: 🌫️ Diffusion Models Explained Like You're 5
发布: 2026年3月12日
📄 English Summary
🌫️ Diffusion Models Explained Like You're 5
Diffusion models generate images by progressively removing noise. The process starts with a clear photo to which static noise is gradually added until it becomes pure noise. The next step involves learning to reverse this process, removing noise step by step to reveal a realistic image. This can be likened to a magic trick, where one begins with random static noise and cleans it up to ultimately display a clear image. The training process of the model involves starting from a clear photo, adding noise progressively until pure noise is reached, thereby learning the transformations at each step.