What Are Diffusion Models?
7 entriesDiffusion models generate data by learning to reverse a gradual noising process: noise is added to training data step by step, and a network learns to remove it, so that sampling starts from pure noise and denoises toward an image, video or other signal. Key ideas include conditioning with text and classifier-free guidance, running diffusion in a compressed latent space (latent diffusion), replacing the U-Net with a transformer (DiT), and flow matching or rectified flow, which learn straighter paths and fewer sampling steps. Distillation methods such as consistency models speed generation further.
Landmark Papers
23 entriesFrom the first diffusion model to flow matching.
Libraries, Code & Tooling
16 entriesRun, train and fine-tune.
Video & Image Models
15 entriesOpen and commercial generators.
Courses
3 entriesLearn to build diffusion models.
Related Pages
10 entriesNeighbouring pages on this site.
Frequently Asked Questions
What is a diffusion model?
A generative model that learns to reverse a gradual noising process, so it can create images, video or audio by starting from noise and denoising step by step.
What is latent diffusion?
Running the diffusion process in the compressed latent space of a VAE-style autoencoder instead of pixel space, which makes high-resolution generation far cheaper. Stable Diffusion is the best-known example.
What is flow matching?
A way to train generative models by learning a velocity field that transports noise to data along straighter paths, used by newer image models such as Stable Diffusion 3 and FLUX for fewer sampling steps.
How do I start building with diffusion models?
Use Hugging Face Diffusers for pipelines and fine-tuning, ComfyUI for visual workflows, and the Hugging Face and DeepLearning.AI courses to learn the internals.
How current is this list?
Links were checked in October 2026. Archived projects are labelled; sites that block automated checks are kept and may need a manual click.