The model was trained on images obtained from the Midjourney model. Try to pull out the visual style of Midjourney and create your own stable diffusion model based on it.