{"paper":{"title":"MVDream: Multi-view Diffusion for 3D Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A multi-view diffusion model trained on both 2D and 3D data acts as a generalizable 3D prior that improves consistency in text-to-3D generation.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Jianglong Ye, Kejie Li, Mai Long, Peng Wang, Xiao Yang, Yichun Shi","submitted_at":"2023-08-31T07:49:06Z","abstract_excerpt":"We introduce MVDream, a diffusion model that is able to generate consistent multi-view images from a given text prompt. Learning from both 2D and 3D data, a multi-view diffusion model can achieve the generalizability of 2D diffusion models and the consistency of 3D renderings. We demonstrate that such a multi-view diffusion model is implicitly a generalizable 3D prior agnostic to 3D representations. It can be applied to 3D generation via Score Distillation Sampling, significantly enhancing the consistency and stability of existing 2D-lifting methods. It can also learn new concepts from a few 2"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We demonstrate that such a multi-view diffusion model is implicitly a generalizable 3D prior agnostic to 3D representations. It can be applied to 3D generation via Score Distillation Sampling, significantly enhancing the consistency and stability of existing 2D-lifting methods.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That joint training on 2D and 3D data produces a prior that remains generalizable to novel text prompts and 3D shapes without overfitting to the specific 3D renderings used or sacrificing single-view quality.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MVDream is a multi-view diffusion model that functions as a generalizable 3D prior, enabling more consistent text-to-3D generation and few-shot 3D concept learning from 2D examples.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A multi-view diffusion model trained on both 2D and 3D data acts as a generalizable 3D prior that improves consistency in text-to-3D generation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"4d472f64f3c14e59151f739b6333b0fef61973fa7492b6460feae3086ce71405"},"source":{"id":"2308.16512","kind":"arxiv","version":4},"verdict":{"id":"0a18a511-cce3-431b-b005-758833fd1b28","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T08:31:50.509985Z","strongest_claim":"We demonstrate that such a multi-view diffusion model is implicitly a generalizable 3D prior agnostic to 3D representations. It can be applied to 3D generation via Score Distillation Sampling, significantly enhancing the consistency and stability of existing 2D-lifting methods.","one_line_summary":"MVDream is a multi-view diffusion model that functions as a generalizable 3D prior, enabling more consistent text-to-3D generation and few-shot 3D concept learning from 2D examples.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That joint training on 2D and 3D data produces a prior that remains generalizable to novel text prompts and 3D shapes without overfitting to the specific 3D renderings used or sacrificing single-view quality.","pith_extraction_headline":"A multi-view diffusion model trained on both 2D and 3D data acts as a generalizable 3D prior that improves consistency in text-to-3D generation."},"references":{"count":161,"sample":[{"doi":"","year":2023,"title":"https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0","work_id":"bca32d14-ca0e-45d5-aabb-29735f777c42","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"https://sketchfab.com/3d-models/popular","work_id":"b2699ee7-f789-4b50-968f-62e0fbf6f7a7","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"https://huggingface.co/DeepFloyd","work_id":"0804d3f6-1aaa-4ec4-a0fc-cbc41b1f14d6","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"https://lumalabs.ai/dashboard/imagine","work_id":"8ceb3039-f241-407f-b9d8-ce4b2e13ee17","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"https://huggingface.co/spaces/lambdalabs/stable-diffusion-image-variations","work_id":"5faab264-b3c9-45c5-80eb-71dc59ff0c18","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":161,"snapshot_sha256":"e6b46909af3c518bf1f71bab7612cb73cd6eb27d5d6f6c85a75393bc270793e4","internal_anchors":10},"formal_canon":{"evidence_count":2,"snapshot_sha256":"9909097488ee0cf5b31d5b3c6738472d570a3600b5c45167e05e0998051ec760"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}