{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZHXVBTKHPSFCHY4YHVUPKM3H65","short_pith_number":"pith:ZHXVBTKH","schema_version":"1.0","canonical_sha256":"c9ef50cd477c8a23e3983d68f53367f77340812cd5cd1dae56ce399b4bda4b14","source":{"kind":"arxiv","id":"2605.14136","version":1},"attestation_state":"computed","paper":{"title":"TeDiO: Temporal Diagonal Optimization for Training-Free Coherent Video Diffusion","license":"http://creativecommons.org/licenses/by/4.0/","headline":"TeDiO improves temporal coherence in video diffusion by smoothing irregular diagonals in self-attention maps.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Gedas Bertasius, Heather Yu, Marc Niethammer, Nurislam Tursynbek, Zhiqiang Lao","submitted_at":"2026-05-13T21:39:50Z","abstract_excerpt":"Recent text-to-video diffusion transformers generate visually compelling frames, yet still struggle with temporal coherence, often producing flickering, drifting, or unstable motion. We show that these failures leave a clear imprint inside the model: incoherent videos consistently exhibit irregular, fragmented temporal diagonals in their intermediate self-attention maps, whereas stable motion corresponds to smooth, band-diagonal patterns. Building on this observation, we introduce TeDiO, a training-free, inference-time method that reinforces temporal consistency by regularizing these internal "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.14136","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T21:39:50Z","cross_cats_sorted":[],"title_canon_sha256":"2df332f88e4569fb9628aa4afdb41d113ad09be1c3d13ac421b83d85364364dd","abstract_canon_sha256":"7453587843dc047d4d68602196bb7bc5f8a38c218cb77f8b22f6c40537b0e7e6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:11.739297Z","signature_b64":"QNfV/9lnJZzrKdO4f63Q0mOO+oV0K71/s19GBedCtH2JkNHR+lAzHT+WEz45uJNBqdZO1uexeyZRPnTtkuDvBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c9ef50cd477c8a23e3983d68f53367f77340812cd5cd1dae56ce399b4bda4b14","last_reissued_at":"2026-05-17T23:39:11.738810Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:11.738810Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TeDiO: Temporal Diagonal Optimization for Training-Free Coherent Video Diffusion","license":"http://creativecommons.org/licenses/by/4.0/","headline":"TeDiO improves temporal coherence in video diffusion by smoothing irregular diagonals in self-attention maps.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Gedas Bertasius, Heather Yu, Marc Niethammer, Nurislam Tursynbek, Zhiqiang Lao","submitted_at":"2026-05-13T21:39:50Z","abstract_excerpt":"Recent text-to-video diffusion transformers generate visually compelling frames, yet still struggle with temporal coherence, often producing flickering, drifting, or unstable motion. We show that these failures leave a clear imprint inside the model: incoherent videos consistently exhibit irregular, fragmented temporal diagonals in their intermediate self-attention maps, whereas stable motion corresponds to smooth, band-diagonal patterns. Building on this observation, we introduce TeDiO, a training-free, inference-time method that reinforces temporal consistency by regularizing these internal "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across multiple video diffusion models (e.g., Wan2.1, CogVideoX), TeDiO delivers markedly smoother motion while preserving per-frame visual quality.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That irregular, fragmented temporal diagonals in self-attention maps are the primary cause of temporal incoherence and that lightweight latent updates to promote diagonal smoothness will reliably fix it without introducing new artifacts or degrading quality.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"TeDiO regularizes temporal diagonals in diffusion transformer attention maps to produce smoother video motion while keeping per-frame quality intact.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"TeDiO improves temporal coherence in video diffusion by smoothing irregular diagonals in self-attention maps.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"46464cd42366f7df8c707069afd0c526c3e60f1160afa85d6f73ca8ecf6ad136"},"source":{"id":"2605.14136","kind":"arxiv","version":1},"verdict":{"id":"e066b3b1-6cf2-4e05-84cb-be40f3596398","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T04:56:27.059297Z","strongest_claim":"Across multiple video diffusion models (e.g., Wan2.1, CogVideoX), TeDiO delivers markedly smoother motion while preserving per-frame visual quality.","one_line_summary":"TeDiO regularizes temporal diagonals in diffusion transformer attention maps to produce smoother video motion while keeping per-frame quality intact.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That irregular, fragmented temporal diagonals in self-attention maps are the primary cause of temporal incoherence and that lightweight latent updates to promote diagonal smoothness will reliably fix it without introducing new artifacts or degrading quality.","pith_extraction_headline":"TeDiO improves temporal coherence in video diffusion by smoothing irregular diagonals in self-attention maps."},"references":{"count":56,"sample":[{"doi":"","year":null,"title":"Prolific.https://www.prolific.com/. 7","work_id":"2f65ca33-e999-40b9-8d76-44a72a9119c4","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Cross-image attention for zero- shot appearance transfer","work_id":"7e822769-42c8-475a-92b3-91ae526d6188","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Uniedit: A unified tuning- free framework for video motion and appearance editing","work_id":"daa0c536-ddbd-41ae-8a50-3447fcf0c0d8","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Separate-and-enhance: Composi- tional finetuning for text-to-image diffusion models","work_id":"1d368525-d098-4165-9d22-b555d6a0b449","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Cd- tvd: Contrastive diffusion for 3d super-resolution with scarce high-resolution time-varying data.arXiv preprint arXiv:2508.08173, 2025","work_id":"769afac9-dc01-475e-b27c-ecf4246f40f0","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":56,"snapshot_sha256":"b18a97f71c400d68825b225c97c39d8f5bfea0e54a6577ff6a17ed2352971bdf","internal_anchors":14},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.14136","created_at":"2026-05-17T23:39:11.738886+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.14136v1","created_at":"2026-05-17T23:39:11.738886+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14136","created_at":"2026-05-17T23:39:11.738886+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZHXVBTKHPSFC","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZHXVBTKHPSFCHY4Y","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZHXVBTKH","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZHXVBTKHPSFCHY4YHVUPKM3H65","json":"https://pith.science/pith/ZHXVBTKHPSFCHY4YHVUPKM3H65.json","graph_json":"https://pith.science/api/pith-number/ZHXVBTKHPSFCHY4YHVUPKM3H65/graph.json","events_json":"https://pith.science/api/pith-number/ZHXVBTKHPSFCHY4YHVUPKM3H65/events.json","paper":"https://pith.science/paper/ZHXVBTKH"},"agent_actions":{"view_html":"https://pith.science/pith/ZHXVBTKHPSFCHY4YHVUPKM3H65","download_json":"https://pith.science/pith/ZHXVBTKHPSFCHY4YHVUPKM3H65.json","view_paper":"https://pith.science/paper/ZHXVBTKH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.14136&json=true","fetch_graph":"https://pith.science/api/pith-number/ZHXVBTKHPSFCHY4YHVUPKM3H65/graph.json","fetch_events":"https://pith.science/api/pith-number/ZHXVBTKHPSFCHY4YHVUPKM3H65/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZHXVBTKHPSFCHY4YHVUPKM3H65/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZHXVBTKHPSFCHY4YHVUPKM3H65/action/storage_attestation","attest_author":"https://pith.science/pith/ZHXVBTKHPSFCHY4YHVUPKM3H65/action/author_attestation","sign_citation":"https://pith.science/pith/ZHXVBTKHPSFCHY4YHVUPKM3H65/action/citation_signature","submit_replication":"https://pith.science/pith/ZHXVBTKHPSFCHY4YHVUPKM3H65/action/replication_record"}},"created_at":"2026-05-17T23:39:11.738886+00:00","updated_at":"2026-05-17T23:39:11.738886+00:00"}