Michelangelo
Conditional 3D shape generation from images or text via a shape-image-text aligned latent space.
About
Michelangelo, a NeurIPS 2023 paper from Tencent PCG and ShanghaiTech, takes an alignment-before-generation approach to a genuinely hard problem: making 3D shapes that actually match a text or image condition. It first encodes shapes into a latent space aligned with image and text embeddings (the SITA-VAE), then learns a conditional diffusion model (ASLDM) that maps from the condition to that latent space. The claim is shapes that semantically conform to the input — a chair described as a chair comes out chair-shaped — rather than the inconsistent outputs the authors say direct conditional training produces.
The code is on GitHub under GPL-3.0 with just under 500 stars, and weights are downloadable from a community Hugging Face mirror (not an official release). There is no formal benchmark table; evaluation in the paper is qualitative and comparative. The copyleft GPL-3.0 license is worth noting for anyone embedding it in a commercial product.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
No published benchmark scores; evaluated qualitatively on shape-image-text alignment in the paper
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: peer-reviewed at NeurIPS 2023; code public under GPL-3.0 (~490 stars); weights available via a community HF mirror. Strengths: novel alignment-before-generation formulation, peer-reviewed, code open. Docked for: (1) weights are a third-party mirror, not an official release; (2) no quantitative benchmarks, only qualitative evaluation; (3) GPL-3.0 copyleft limits commercial reuse; (4) modest community adoption. License: code GPL-3.0; weights licensing unclear.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 8.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · NeurIPS 2023 (Tencent PCG, ShanghaiTech)
- · Code GPL-3.0, ~490 stars; weights via third-party HF mirror
- · No quantitative benchmarks; qualitative evaluation only
- · Copyleft license limits commercial reuse