A blog related to original AI artworks. I make AI artworks using coined words and emoji as prompts.

AI's Creation of Various Images by Learning from Kandinsky, “Negating” Utrillo, and Changing the Scale

By providing Kandinsky's name as both a positive and negative prompt to Stable Diffusion 2.1 and further adjusting the “scale,” various images with different styles were obtained.

By providing Kandinsky's name in various forms, a surprisingly diverse range of images can be created. However, in this case, providing Utrillo's name as a negative prompt resulted in even more interesting images. One of the parameters for adjusting the generated images is the “scale,” which, on OpenAI.org, can be varied within the range of 2 to 30. Typically, a value of 8 or lower is used, but in this case, much finer images are generated.

This includes specifying names of artists other than Kandinsky as positive prompts as well. Normally, setting the scale to around 30 results in distorted images, but in this case, it produces images with a pop art style.