I did everything manually with stable diffusion, using the "img2img" function that allows you to use an image as base and change it keeping a certain level of consistency, more or less.
