Character consistency via cached SFace + DINOv2 embeddings, no LoRA

A Stable Diffusion user's technique combines SFace + DINOv2 to cache character reference embeddings in a portable .char file, achieving character consistency without training a LoRA. It builds on Facebook's DINOv2, a vision model trained without supervision.