Unique3D
High-fidelity textured mesh generation from a single image — multi-view and normal diffusion with a 30-second ISOMER reconstruction.
About
Unique3D is built for one thing: squeezing the most geometric and textural detail out of a single input image. It runs a multi-view diffusion model alongside a dedicated normal diffusion model, progressively upscales the generated views for finer resolution, then hands everything to ISOMER, an instant mesh reconstruction algorithm that bakes color and geometry priors directly into the final mesh. From upload to textured mesh in roughly 30 seconds.
The work appeared at NeurIPS 2024 and, per the authors' experiments, beats other image-to-3D baselines on geometric and textural detail. It's fully open source under MIT with weights on Hugging Face, and the repo even has ComfyUI and Windows support. It does have a quirk worth knowing: the mesh is normalized by the longest edge of the object, so feeding it an image where the object's longest edge is visible (ideally a front-facing rest pose) gives noticeably better results.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Paper (arXiv 2405.20343): reports outperforming image-to-3D baselines on geometric and textural detail
- NeurIPS 2024; no headline benchmark numbers independently verified
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 (TIER verified, metrics-v1). Evidence: NeurIPS 2024 paper (arXiv 2405.20343) with experiments against image-to-3D baselines; MIT-licensed official repo (AiuniAI/Unique3D) with ~3.6k stars; weights on Hugging Face; project page live. Strengths: peer-reviewed, permissive license, strong community uptake. Docked for: (1) headline claims are author-reported; (2) output quality depends on input pose and occlusion (per the authors' own tips). License: MIT.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · NeurIPS 2024 paper (arXiv 2405.20343), authors at Tsinghua University
- · Official repo AiuniAI/Unique3D, MIT, ~3.6k stars, weights on Hugging Face (Aipico/Unique3D)
- · 30-second ISOMER reconstruction; quality sensitive to input pose per authors' own tips