Made using stable diffusion.

• Initial image was just an abstract black and white blob which was then sent to img2img, I find doing this can make the composition more interesting.
• img2img using Dreamlike Photoreal model + some custom trained embedding. Then upscaled and inpainted details using other models.
• The AI tends to be very bad at faces that aren’t vertical, so I had to rotate the whole image by 60 degrees to work on the face.