{"paper":{"title":"Ovi: Twin Backbone Cross-Modal Fusion for Audio-Video Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Ovi generates audio and video together in one process by fusing twin identical diffusion transformers block by block.","cross_cats":["cs.CV","cs.SD","eess.AS"],"primary_cat":"cs.MM","authors_text":"Calder Katyal, Chetwin Low, Weimin Wang","submitted_at":"2025-09-30T21:03:50Z","abstract_excerpt":"Audio-video generation has often relied on complex multi-stage architectures or sequential synthesis of sound and visuals. We introduce Ovi, a unified paradigm for audio-video generation that models the two modalities as a single generative process. By using blockwise cross-modal fusion of twin-DiT modules, Ovi achieves natural synchronization and removes the need for separate pipelines or post hoc alignment. To facilitate fine-grained multimodal fusion modeling, we initialize an audio tower with an architecture identical to that of a strong pretrained video model. Trained from scratch on hund"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"By using blockwise cross-modal fusion of twin-DiT modules, Ovi achieves natural synchronization and removes the need for separate pipelines or post hoc alignment.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That initializing an audio tower with an architecture identical to a strong pretrained video model and then jointly training via bidirectional cross-attention on a vast video corpus will produce reliable cross-modal timing and semantic alignment without additional alignment losses or post-processing.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A single generative model uses twin DiT backbones with blockwise cross-attention and scaled-RoPE timing exchange to synthesize synchronized audio-video directly.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Ovi generates audio and video together in one process by fusing twin identical diffusion transformers block by block.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0a6ffc5657c0aa9898b238dbae0f54c6ad4605fe468ffdffa0f699d84adce477"},"source":{"id":"2510.01284","kind":"arxiv","version":1},"verdict":{"id":"cd90296c-aac0-4d45-af8b-ff608b2ecbbf","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T00:58:33.434002Z","strongest_claim":"By using blockwise cross-modal fusion of twin-DiT modules, Ovi achieves natural synchronization and removes the need for separate pipelines or post hoc alignment.","one_line_summary":"A single generative model uses twin DiT backbones with blockwise cross-attention and scaled-RoPE timing exchange to synthesize synchronized audio-video directly.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That initializing an audio tower with an architecture identical to a strong pretrained video model and then jointly training via bidirectional cross-attention on a vast video corpus will produce reliable cross-modal timing and semantic alignment without additional alignment losses or post-processing.","pith_extraction_headline":"Ovi generates audio and video together in one process by fusing twin identical diffusion transformers block by block."},"references":{"count":20,"sample":[{"doi":"","year":null,"title":"Seed-TTS: A Family of High-Quality Versatile Speech Generation Models","work_id":"6e88ee95-1133-4302-a142-cdf8f9456a8d","ref_index":1,"cited_arxiv_id":"2406.02430","is_internal_anchor":true},{"doi":"","year":2025,"title":"com/research/video-generation-models-as-world-simulators","work_id":"e01a4ca3-169f-41c9-adf9-19a9dd64a994","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"HuMo: Human-centric video generation via collaborative multi-modal conditioning","work_id":"313111a9-b492-44f5-bf1f-6b5c2c642abf","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"CosyVoice: A Scalable Multilingual Zero-shot Text-to-speech Synthesizer based on Supervised Semantic Tokens","work_id":"e5ad925a-4045-49b5-b301-208bcbf3eca8","ref_index":4,"cited_arxiv_id":"2407.05407","is_internal_anchor":true},{"doi":"","year":null,"title":"Ace-step: A step towards music generation foundation model.arXiv preprint arXiv:2506.00045","work_id":"0d4831b4-71c3-4405-b540-1683d26109f6","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":20,"snapshot_sha256":"82668244fa647f84970ed0b7ad60dd132801c0e8ed2357f8d3cfae7048cb0fe5","internal_anchors":6},"formal_canon":{"evidence_count":2,"snapshot_sha256":"dcc63dfbc2e0470515064e223638b706ac28e96ca692fa011a4f85b128f3a7aa"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}