Variance Reduction for Diffusion Teachers Improves Text-to-3D and Distillation Pipelines

arXiv·low signal

Paper presents variance reduction techniques for downstream pipelines that use pretrained diffusion models as frozen teachers, directly applicable to text-to-3D generation and single-step distillation workflows. For practitioners building diffusion-based generation pipelines, this offers a practical method to reduce training variance without modifying the teacher model.