Kandinsky as You Preferred

Due to the significant generative capabilities of GenAI (generative artificial intelligence), the art community has actively embraced it to create painterly content. Large text-to-image models can quickly generate aesthetically pleasing outcomes. However, the process can be non-deterministic and often involves tedious trial-and-error as users struggle to formulate effective prompts to achieve their desired results. This paper describes a generative approach that empowers users to easily work with a large text-to-image (TTI) model to create their preferred painterly content. The authors propose a large model personalization method, namely Semantic Injection, to personalize a large TTI model in a given specific artistic style, i.e., Kandinsky’s paintings in Bauhaus era, as the Artist Model. Through working with a Kandinsky expert, the authors first establish a semantic descriptive guideline and a TTI dataset of Kandinsky style and then apply the Semantic Injection method to obtain an Artist Model of Kandinsky, empowering users to create preferred Kandinsky content in a deterministically controllable manner.