DreamCraft3D generates 3D content by leveraging a 2D reference image to guide geometry sculpting and texture boosting stages. It uses view-dependent diffusion models for score distillation sampling to ensure geometry consistency, and employs a bootstrapped score distillation approach where a personalized Dreambooth model is trained on augmented scene renderings to provide view-consistent texture guidance. The system alternates between optimizing the diffusion prior and the 3D scene representation for mutually reinforcing improvements.