No prompt

Multiple galleries featuring images created using various Stable Diffusion models (SDXL fine tuned model from civitai.com), using an empty text-to-image prompt.

Each gallery showcases a unique selection of visuals generated solely by altering the (random) seed. This not only provides a diverse array of imagery but also serves as a means to underscore the inherent biases of different models. Notably, when prompted with emptiness, certain styles tend to generate a disproportionate number of specific types of images.

It’s not a scientific experiment (the number of images generated is too small), but it’s still representative of the biases present in the datasets used to train the different models.