Diffusion Models
Diffusion models
Most image generation today runs on diffusion models, so the first need is a clear picture of what they are. This part gives two conceptual explainers; the second part below adds guided introductions, a build-your-own tutorial, and one fine-tuned model.
The same notes are in Diffusion Models.
Lilian Weng on diffusion is her post "What are Diffusion Models?", the reference explainer to start from.
Diffusion models made easy — J. Rafid. It places diffusion after GANs and VAEs: GANs work for many applications but are hard to train and lose diversity through mode collapse and vanishing gradients, while VAEs have the more solid theoretical foundation, and diffusion is introduced against those two.
Diffusion Models
Once the idea is clear, the next step is to see it built and then used. The list runs from an introduction, to a minimal implementation, to where the ideas came from, and ends with a model you can download.
The same notes are in Diffusion models and GenAI Applications.
Intro to Diffusion Models is AssemblyAI's easy-to-follow guide to diffusion models, one of the biggest developments in machine learning of the past several years.
Build your own is the MinImagen tutorial: a minimal implementation of the Imagen text-to-image model, from the same year as DALL-E 2.
How physics advanced Generative AI is a high-level look at how concepts from physics, thermodynamics among them, inspired many cutting-edge generative models.
Openjourney is an open source Stable Diffusion fine tuned model on Midjourney images, by PromptHero
Deprecated links
These links and images no longer work. The original wording is kept here. A same-resource copy, when one was checked, is used above.
Diffusion models made easy. This address no longer opens: https://towardsdatascience.com/diffusion-models-made-easy-8414298ce4da
Last updated
Was this helpful?