{"paper":{"title":"Structured 3D Latents for Scalable and Versatile 3D Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A structured latent that merges sparse 3D grids with dense multiview features supports high-quality generation of 3D assets in multiple output formats from text or image input.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bowen Zhang, Dong Chen, Jianfeng Xiang, Jiaolong Yang, Ruicheng Wang, Sicheng Xu, Xin Tong, Yu Deng, Zelong Lv","submitted_at":"2024-12-02T13:58:38Z","abstract_excerpt":"We introduce a novel 3D generation method for versatile and high-quality 3D asset creation. The cornerstone is a unified Structured LATent (SLAT) representation which allows decoding to different output formats, such as Radiance Fields, 3D Gaussians, and meshes. This is achieved by integrating a sparsely-populated 3D grid with dense multiview visual features extracted from a powerful vision foundation model, comprehensively capturing both structural (geometry) and textural (appearance) information while maintaining flexibility during decoding. We employ rectified flow transformers tailored for"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our model generates high-quality results with text or image conditions, significantly surpassing existing methods, including recent ones at similar scales.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The integration of a sparsely-populated 3D grid with dense multiview visual features from a vision foundation model comprehensively captures both structural geometry and textural appearance information while maintaining decoding flexibility.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SLAT provides a unified 3D latent representation enabling versatile high-quality generation across multiple output formats from text or image inputs.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A structured latent that merges sparse 3D grids with dense multiview features supports high-quality generation of 3D assets in multiple output formats from text or image input.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ba52f5368bf37445b0ea7ad2f8587a46f42b06282058f3f1780ddaca9cbd6231"},"source":{"id":"2412.01506","kind":"arxiv","version":3},"verdict":{"id":"6e031692-e933-4e9f-be68-994e503b3da1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T15:05:40.004428Z","strongest_claim":"Our model generates high-quality results with text or image conditions, significantly surpassing existing methods, including recent ones at similar scales.","one_line_summary":"SLAT provides a unified 3D latent representation enabling versatile high-quality generation across multiple output formats from text or image inputs.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The integration of a sparsely-populated 3D grid with dense multiview visual features from a vision foundation model comprehensively captures both structural geometry and textural appearance information while maintaining decoding flexibility.","pith_extraction_headline":"A structured latent that merges sparse 3D grids with dense multiview features supports high-quality generation of 3D assets in multiple output formats from text or image input."},"references":{"count":117,"sample":[{"doi":"","year":2024,"title":"Gpt-4o system card. 2024. 6, 16","work_id":"31f4f27a-bda9-4fa5-9486-8dd8a76923d3","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","ref_index":2,"cited_arxiv_id":"2303.08774","is_internal_anchor":true},{"doi":"","year":null,"title":"Build- ing normalizing flows with stochastic interpolants","work_id":"18c42973-7784-4a94-9eb5-a5c41d8ae522","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Improving image generation with better captions","work_id":"283b326a-614b-42d1-b67f-4040cddad7e5","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Demystifying mmd gans","work_id":"d3a22946-30b0-4329-896b-dff78f4495e1","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":117,"snapshot_sha256":"3f8bfa1c68f0548796791686cf0bdacb503eaa262650f2c63bde43a220401fbc","internal_anchors":6},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2fb8040140c7f52b28f5d73f2ea22f598eb5d2c2696eb6765c1ff4ef87346ac1"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}