About ashawkey/stable-dreamfusion

The widely used open reimplementation of DreamFusion: Text-to-3D using 2D Diffusion (Poole et al., 2022) by Jiaxiang Tang. See the paper page for context.

Because the original DreamFusion used Google's closed Imagen diffusion model, this repo recreates the pipeline on top of Stable Diffusion, providing reference implementations of Score Distillation Sampling (SDS), a hash-grid backed NeRF/DMTet geometry representation, and the text-to-3D optimization loop. It is the de-facto starting point that Magic3D, ProlificDreamer, and Magic123 in the corpus extend. Mirrored under Apache-2.0 — README only.