ID-to-3D is an innovative method that can generate 3D head models with identity and text guidance from a random, everyday photo. It features separable expressions. This method is combinatorial, using a task-specific 2D diffusion model as an optimized prior. By extending the base model and adding lightweight expression and identity-aware architectures, a 2D prior is created for geometry and texture generation, and refined by fine-tuning only 0.2% of the available training parameters. Combining powerful facial identity embeddings and neural representations, this method not only accurately reconstructs facial features but also reconstructs accessories and hair, and provides render-ready assets suitable for games and remote presentations.