Diffusion Models

Diffusion models generate data by learning to reverse a gradual noising process. This page keeps the core diffusion literature separate from text-to-image system reports and flow-matching/fast-sampling papers.

Surveys and Monographs

Year	Paper	Topic	Note
2025	The Principles of Diffusion Models	Monograph / survey	Unifies variational, score-based, and flow-based views of diffusion models.

Origins and Core Theory

Year	Paper	Topic	Note
2015	Deep Unsupervised Learning using Nonequilibrium Thermodynamics	Early diffusion	Early noising-reversal view of generative modeling.
2019	Generative Modeling by Estimating Gradients of the Data Distribution	NCSN	Noise-conditioned score networks for score-based generation.
2020	Denoising Diffusion Probabilistic Models (code)	DDPM	Modern denoising diffusion formulation.
2020	Denoising Diffusion Implicit Models	DDIM	Faster non-Markovian sampling with DDPM training.
2020	Score-Based Generative Modeling through Stochastic Differential Equations	Score-based SDEs	Continuous-time unification of diffusion and score models.
2021	Improved Denoising Diffusion Probabilistic Models (code)	Improved DDPM	Learned variances and improved sampling/likelihoods.

Guidance, Latents, and Architecture

Year	Paper	Topic	Note
2021	Diffusion Models Beat GANs on Image Synthesis (code)	Classifier guidance	Shows diffusion can beat GANs in image quality with improved architectures and guidance.
2021	High-Resolution Image Synthesis with Latent Diffusion Models (code)	Latent diffusion	Diffusion in autoencoder latent space; Stable Diffusion lineage.
2022	Classifier-Free Diffusion Guidance	Guidance	Conditional guidance without an external classifier.
2022	Elucidating the Design Space of Diffusion-Based Generative Models	EDM	Modular analysis of diffusion design choices.
2022	Scalable Diffusion Models with Transformers	DiT	Replaces U-Net backbones with scalable diffusion transformers.
2023	Adding Conditional Control to Text-to-Image Diffusion Models	ControlNet	Adds spatial controls to pretrained text-to-image diffusion models.

Cross-Database Pointers

Theme	Go To	Note
Text-to-image systems	Text-to-Image and Video Systems	GLIDE, DALL-E 2/3, Imagen, SDXL, Sora, and related system papers live there.
Fast sampling and flow-based successors	Flow Matching and Fast Sampling	DPM-Solver, consistency models, rectified flow, flow matching, and Stable Diffusion 3 live there.

Reading Path

Step	Read
1	The Principles of Diffusion Models for the unified conceptual map.
2	Nonequilibrium Thermodynamics, NCSN, and DDPM.
3	DDIM, Score-Based SDEs, and Improved DDPM.
4	Guided Diffusion, Classifier-Free Guidance, and EDM.
5	Latent Diffusion, DiT, and ControlNet.