DreamFusion
The 2022 Google paper that invented score distillation sampling and kicked off text-to-3D.
About
DreamFusion showed that a frozen text-to-image diffusion model could supervise the optimization of a Neural Radiance Field, producing a 3D object from a plain caption with no 3D training data at all. Its core invention, score distillation sampling, treats the diffusion model as a critic that scores random-angle renderings until the scene matches the prompt, and the resulting NeRFs are coherent, relightable, and exportable to meshes.
Published in September 2022 by Ben Poole, Ajay Jain, Jonathan Barron, and Ben Mildenhall, it became the most influential text-to-3D paper of its era and the template for everything that followed. Google never released code, weights, or an API, so the community reimplemented it in projects like threestudio; this entry covers the method itself.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
No published benchmarks; qualitative gallery on the project page
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: landmark arXiv paper (Sep 2022) from Google Research and UC Berkeley with thousands of citations; an entire subfield was built on its SDS method. Strengths: foundational method, fully described in the paper. Docked for: (1) no official code, weights, or API ever released; (2) per-prompt optimization takes about 1.5 hours, obsolete as a practical tool. License: research publication, no license offered.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 3.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Google Research + UC Berkeley; arXiv Sep 2022; invented score distillation sampling
- · Most-cited text-to-3D paper of its era; template for the whole subfield
- · No published benchmarks; community reimplementations (threestudio) only
- · Per-prompt NeRF optimization ~1.5h; obsolete as a practical tool