People who've been around a while may remember my love of gibberish prompts - prompts made from combining multiple different writing systems in ways which shouldn't make sense in any human language. This was always particularly fun in the early Stable Diffusion models and SDXL 1.0. Well, it looks like it's gonna be fun for me in some newer models too! In this model, long gibberish seems to work better here than short gibberish, and adding "No text" to the prompt is generally a good idea.