Era3D
NeurIPS 2024 multiview diffusion that turns one photo into six consistent 512px views, then a detailed mesh.
About
Era3D generates high-resolution multiview images from a single input photo and reconstructs a textured 3D mesh from them. Its two tricks are a diffusion-based camera predictor that estimates focal length and elevation so shapes do not distort, and a row-wise attention layer that enforces epipolar geometry while cutting multiview attention compute roughly 12x versus dense attention. The result is six consistent 512x512 views where earlier methods topped out lower.
Published at NeurIPS 2024 by researchers at HKUST, HKU, and DreamTech, with code and weights released. The license is AGPL-3.0, which is unusually strict for this space: anything built on the code or weights must itself be open-sourced, so commercial teams need to read the terms carefully.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
12x lower multiview-attention compute vs dense attention at 512px (author-reported, paper); no independent evals
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: NeurIPS 2024 peer-reviewed paper (arXiv:2405.11616) with official code and weights published. Strengths: peer-reviewed, genuinely novel efficiency angle, reproducible. Docked for: (1) AGPL-3.0 license forces open-sourcing of derivatives, limiting commercial adoption; (2) modest community traction versus newer image-to-3D models. License: AGPL-3.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · HKUST/HKU/DreamTech; NeurIPS 2024 (arXiv:2405.11616)
- · Single image to 6 consistent 512px views + textured mesh; row-wise attention ~12x compute cut (author-reported)
- · ~644 stars; modest community traction vs newer image-to-3D models
- · No independent evals; AGPL limits commercial adoption