LMU researchers release Stable Diffusion AI model for image synthesis
Researchers at Ludwig-Maximilians-Universität München have developed an AI model capable of generating images from text prompts using standard consumer hardware.
The Machine Vision & Learning group at Ludwig-Maximilians-Universität (LMU) München, led by Professor Björn Ommer, has developed an AI algorithm named Stable Diffusion. The model enables users to generate complex images from simple text descriptions without the need for supercomputing infrastructure.
By condensing the features of billions of training images into a file size of just a few gigabytes, the team achieved a model that runs on standard graphics cards. This advancement was supported by the startup Stability.Ai, which provided the server capacity necessary for training the algorithm. The researchers view this as a significant step in making machine learning tools more accessible for both amateur users and professional artists.
The project aims to advance basic research in computer vision and machine learning. To support further academic and practical innovation, the LMU team has released the model under the CreativeML Open RAIL-M license, allowing for open access and community collaboration.