What it is
Open 3D game-asset corpus with text-conditioned meshes, contextual safety tiers, and reusable multi-view, SDF, point-cloud, and embedding artifacts. 9.3K text-conditioned game-asset meshes from 9.4K prompts, structured by 14 genres × 29 asset groups with 4 safety tiers; each asset includes watertight mesh, 18 views, SDF samples, 3DGS points, and DINO/CLIP features.
Key results
- 9,307 meshes from 9,419 curated prompts
- 18 calibrated 800×800 views per asset
- Apache 2.0; includes SDF, 3DGS points, DINO and CLIP features
Useful for evaluating autoregressive text-to-3D across genre, asset type, and contextual safety tiers.