Set your Publisher ID in src/config/ads.ts to display live ads.
Diffusion Models & Image Generation
Forward noise addition, reverse denoising, UNets, and Latent Diffusion
Diffusion models generate photorealistic images by learning to reverse a gradual noising process, synthesizing clean images from pure Gaussian static.
The Forward and Reverse Diffusion Process
Forward Process: Progressively adds Gaussian noise to an image over T timesteps until it turns into unrecognizable random noise.
Reverse Process: A neural network (typically a UNet with cross-attention) is trained to predict and subtract the added noise at each timestep t.
Latent Diffusion (Stable Diffusion): Operating in high-resolution pixel space is computationally prohibitive. Latent diffusion uses a Variational Autoencoder (VAE) to compress images into a low-dimensional latent space where denoising occurs.
Classifier-Free Guidance (CFG) scale controls how strictly the diffusion model adheres to the text prompt versus generating diverse artistic variations.
Set your Publisher ID in src/config/ads.ts to display live ads.
Test Yourself With Exercise
In Latent Diffusion Models (such as Stable Diffusion), what is the purpose of compressing images into latent space with a VAE?