{"paper":{"title":"AnyFlow: Any-Step Video Diffusion Model with On-Policy Flow Map Distillation","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"AnyFlow distills video diffusion models to match few-step consistency performance while preserving scaling as sampling steps increase.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Guian Fang, Han Cai, Mike Zheng Shou, Song Han, Weijia Mao, Yuchao Gu, Yuxin Jiang","submitted_at":"2026-05-13T16:06:34Z","abstract_excerpt":"Few-step video generation has been significantly advanced by consistency distillation. However, the performance of consistency-distilled models often degrades as more sampling steps are allocated at test time, limiting their effectiveness for any-step video diffusion. This limitation arises because consistency distillation replaces the original probability-flow ODE trajectory with a consistency-sampling trajectory, weakening the desirable test-time scaling behavior of ODE sampling. To address this limitation, we introduce AnyFlow, the first any-step video diffusion distillation framework based"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"AnyFlow achieves performance matches or surpasses consistency-based counterparts in the few-step regime, while scaling with sampling step budgets.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That learning flow-map transitions over arbitrary intervals via backward simulation will avoid introducing new discretization or exposure biases that undermine the original ODE scaling behavior.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"AnyFlow enables any-step video diffusion by distilling flow-map transitions over arbitrary time intervals with on-policy backward simulation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"AnyFlow distills video diffusion models to match few-step consistency performance while preserving scaling as sampling steps increase.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9c75f9b65d69ed1e88dba66e6ec31e331582e53eda41b72928838a7b621cac0d"},"source":{"id":"2605.13724","kind":"arxiv","version":1},"verdict":{"id":"cbad793b-6c2d-4915-875e-7b25da481366","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:59:06.628346Z","strongest_claim":"AnyFlow achieves performance matches or surpasses consistency-based counterparts in the few-step regime, while scaling with sampling step budgets.","one_line_summary":"AnyFlow enables any-step video diffusion by distilling flow-map transitions over arbitrary time intervals with on-policy backward simulation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That learning flow-map transitions over arbitrary intervals via backward simulation will avoid introducing new discretization or exposure biases that undermine the original ODE scaling behavior.","pith_extraction_headline":"AnyFlow distills video diffusion models to match few-step consistency performance while preserving scaling as sampling steps increase."},"references":{"count":49,"sample":[{"doi":"","year":2025,"title":"Wan: Open and Advanced Large-Scale Video Generative Models","work_id":"ad3ebc3b-4224-46c9-b61d-bcf135da0a7c","ref_index":1,"cited_arxiv_id":"2503.20314","is_internal_anchor":true},{"doi":"","year":2025,"title":"Cosmos World Foundation Model Platform for Physical AI","work_id":"a2dba24c-318d-476a-8b21-4289c265810c","ref_index":2,"cited_arxiv_id":"2501.03575","is_internal_anchor":true},{"doi":"","year":2024,"title":"HunyuanVideo: A Systematic Framework For Large Video Generative Models","work_id":"881efa7e-7e73-4c66-9cc3-2803e551061c","ref_index":3,"cited_arxiv_id":"2412.03603","is_internal_anchor":true},{"doi":"","year":2024,"title":"CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer","work_id":"f38fc088-12aa-4bf4-9ecd-08d3e797ccb7","ref_index":4,"cited_arxiv_id":"2408.06072","is_internal_anchor":true},{"doi":"","year":2024,"title":"LTX-Video: Realtime Video Latent Diffusion","work_id":"cee5c521-3ce9-466e-a035-1e42f89254f4","ref_index":5,"cited_arxiv_id":"2501.00103","is_internal_anchor":true}],"resolved_work":49,"snapshot_sha256":"228d63d418e474663924e75d3124beef7bd8dac02899682a6c7f1518c34ca082","internal_anchors":17},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d1c34df08b0fa4b0112fd367b48cc42740575cd8054dda9b99747bd5ba6e4ba1"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}