
Academic lab · Munich, Germany
CompVisLMU Munich
The Computer Vision and Learning group at LMU Munich, which developed latent diffusion, the research behind Stable Diffusion, and releases its code openly.
Munich, Germany
- Type
- Academic lab
- Based in
- Munich, Germany
- Key papers
- 4
- Featured projects
- 8
About
CompVis is the Computer Vision and Learning group at LMU Munich, formerly the Computer Vision Group at Heidelberg University, led by Prof. Dr. Bjorn Ommer. Its research covers generative visual synthesis, invertible deep models for explainable AI, deep metric and representation learning, and self-supervised learning.
The group is best known for latent diffusion models, presented at CVPR 2022 as a method for high-resolution image synthesis. Source code and models were released on GitHub, and the approach underlies Stable Diffusion.
Newer projects listed on the group's site include SCFlow, TREAD and CleanDIFT. The group also applies its methods to the digital humanities and the life sciences.
Mission and ambitionsThe group pursues fundamental research in computer vision and machine learning, with an emphasis on generative models and interpretable representations. It publishes code and models so that others can build on the work.
Known for
- 01
Latent diffusion models
- 02
The research basis of Stable Diffusion
- 03
Open image synthesis code and models
- 04
Explainable and generative vision research
Milestones
- 2021
Taming Transformers
Combined convolutional and transformer models for high-resolution synthesis, with code released.
- 2022
Latent diffusion models
The group presented high-resolution image synthesis with latent diffusion at CVPR and released code and models on GitHub.
- 2022
Stable Diffusion
The group's latent diffusion work was the foundation for the Stable Diffusion image generator.
- 2024
CleanDIFT
Released a method for extracting diffusion features without added noise.
Open source
Latent text-to-image diffusion model (73,576 stars)
High-resolution image synthesis with latent diffusion models (14,190 stars)
VQGAN and transformers for image synthesis (6,533 stars)
Fast monocular depth estimation with flow matching (764 stars)
Most downloaded model of the group, the Stable Diffusion v1.4 weights.
Safety checker component used by Stable Diffusion pipelines.
- CompVis/cleandiftmodel
Diffusion features without noise, from the CleanDIFT paper.
- CompVis/EDGSspace
Demo space for the EDGS Gaussian splatting method.
Research papers
Notable releases
Openness
Code is published on GitHub and weights on Hugging Face, with licences varying by project.
People
- Bjorn OmmerHead of the Computer Vision and Learning group, LMU Munich
- Robin RombachFirst author of latent diffusion; former group PhD student
LDM first author, listed as a group PhD student on the lab site.
- Patrick EsserCo-author of latent diffusion and Taming Transformers
Maintainer of CompVis repositories including stable-diffusion.
- Andreas BlattmannLatent diffusion co-author; co-founder of Black Forest Labs
Listed from the lab's own pages and public profiles.
Contact and community
The group's website lists its secretary, postal address and open positions on a contact page.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



