In regards to AI images, yes 60-80% AI Gen, with most common things such as hair and finer details being worked on manually. But this most likely changes between image to image. There have been examples before with Sofia being more or less completely done by hand.
In regards to consistency, most likely a trained LoRA, ControlNet reference, use of refiners, standardized prompts and configurations. Then a standardized workflow for the manual work (techniques, brushes and filter methods.)
Dr.Muramasa hinted about giving a workflow example of an image one time without providing any specifics, but I guess this got blocked by Kal 😔