Diffusion Models
A class of generative models that create data by gradually removing noise, reversing a process of adding random noise to training data. The technology behind modern AI image generators.
Diffusion models learn to generate data by reversing a noising process. During training, noise is added to images step by step until they become pure random noise. The model learns to reverse this — starting from noise and gradually denoising into a coherent image guided by a text prompt.
How they work:
- Forward Process: Gradually add noise to an image until it becomes random noise
- Reverse Process: Train the model to denoise step by step, conditioned on a text description
- Inference: Start with random noise and apply the learned reverse process to generate a new image
Diffusion models power today’s leading image generators including Midjourney, DALL-E 3, Stable Diffusion, and Adobe Firefly. Variants like latent diffusion (used by Stable Diffusion) operate in a compressed latent space for efficiency. Beyond images, diffusion extends to video generation, audio synthesis, and molecular design.