{"paper":{"title":"EntityBench: Towards Entity-Consistent Long-Range Multi-Shot Video Generation","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"Explicit per-entity memory maintains character consistency across long gaps in multi-shot video generation where existing methods fail.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Meng Wei, Ruozhen He, Vicente Ordonez, Ziyan Yang","submitted_at":"2026-05-14T17:59:55Z","abstract_excerpt":"Multi-shot video generation extends single-shot generation to coherent visual narratives, yet maintaining consistent characters, objects, and locations across shots remains a challenge over long sequences. Existing evaluations typically use independently generated prompt sets with limited entity coverage and simple consistency metrics, making standardized comparison difficult. We introduce EntityBench, a benchmark of 140 episodes (2,491 shots) derived from real narrative media, with explicit per-shot entity schedules tracking characters, objects, and locations simultaneously across easy / medi"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments show that cross-shot entity consistency degrades sharply with recurrence distance in existing methods, and that explicit per-entity memory yields the highest character fidelity (Cohen's d = +2.33) and presence among methods evaluated.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The entity schedules extracted from real narrative media and the fidelity gate used for cross-shot scoring accurately reflect the consistency challenges faced by current video generation models.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"EntityBench is a new benchmark with detailed per-shot entity schedules from real media, and the EntityMem baseline using persistent per-entity memory achieves the highest character fidelity with Cohen's d of +2.33.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Explicit per-entity memory maintains character consistency across long gaps in multi-shot video generation where existing methods fail.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"35bb8b75321305d5184577536cc9736b67cb7b05fe9916729bd8e6d449408a62"},"source":{"id":"2605.15199","kind":"arxiv","version":1},"verdict":{"id":"e0c0b10c-aa4b-4654-a2a7-fea1ceb01554","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T03:10:57.117220Z","strongest_claim":"Experiments show that cross-shot entity consistency degrades sharply with recurrence distance in existing methods, and that explicit per-entity memory yields the highest character fidelity (Cohen's d = +2.33) and presence among methods evaluated.","one_line_summary":"EntityBench is a new benchmark with detailed per-shot entity schedules from real media, and the EntityMem baseline using persistent per-entity memory achieves the highest character fidelity with Cohen's d of +2.33.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The entity schedules extracted from real narrative media and the fidelity gate used for cross-shot scoring accurately reflect the consistency challenges faced by current video generation models.","pith_extraction_headline":"Explicit per-entity memory maintains character consistency across long gaps in multi-shot video generation where existing methods fail."},"references":{"count":34,"sample":[{"doi":"","year":null,"title":"Mixture of contexts for long video generation","work_id":"b800ec2a-ab41-48e3-a98d-a26b4942bb2a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"SkyReels-V2: Infinite-length Film Generative Model","work_id":"2ce11350-273e-4f0d-ae78-292aa3151060","ref_index":2,"cited_arxiv_id":"2504.13074","is_internal_anchor":true},{"doi":"","year":null,"title":"Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities","work_id":"008df105-2fdd-45d8-857a-8e35868aecb6","ref_index":3,"cited_arxiv_id":"2507.06261","is_internal_anchor":true},{"doi":"","year":null,"title":"Narrlv: Towards a comprehensive narrative-centric evaluation for long video generation.arXiv preprint arXiv:2507.11245,","work_id":"b79202ec-ba64-4d4f-b95e-af6e30e82a4a","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Longvie: Multimodal-guided controllable ultra-long video generation","work_id":"3f0217fc-ba5b-4c51-a512-eb2300699c36","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":34,"snapshot_sha256":"d0d450de56368ede9848d269eaf894c7eff4050e0b1218171f67832999c978cf","internal_anchors":9},"formal_canon":{"evidence_count":2,"snapshot_sha256":"0b4ee4d44b3050f2a38465a68565b01c84409343b3d73d9f4762f016ad8284e4"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}