I HAVE AN IDEA!
Why not train a Lora on the ID images and have an ID generating stable diffusion workflow?