{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HEHLHZYFGMTMQKCKHTSP3CG7CA","short_pith_number":"pith:HEHLHZYF","schema_version":"1.0","canonical_sha256":"390eb3e7053326c8284a3ce4fd88df10122f6b796dad200e21c6e3b1dec204a8","source":{"kind":"arxiv","id":"2604.17625","version":2},"attestation_state":"computed","paper":{"title":"FlowC2S: Flowing from Current to Succeeding Frames for Fast and Memory-Efficient Video Continuation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"FlowC2S flows directly from current video frames to succeeding ones, halving input size and outperforming prior methods with five evaluations.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Christian Sandor, Hovhannes Margaryan, Quentin Bammey","submitted_at":"2026-04-19T21:38:21Z","abstract_excerpt":"This paper introduces a novel methodology for generating fast and memory-efficient video continuations. Our method, dubbed FlowC2S, fine-tunes a pre-trained text-to-video flow model to learn a vector field between the current and succeeding video chunks. Two design choices are key. First, we introduce inherent optimal couplings, utilizing temporally adjacent video chunks during training as a practical proxy for true optimal couplings, resulting in straighter flows. Second, we incorporate target inversion, injecting the inverted latent of the target chunk into the input representation to streng"},"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":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2604.17625","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-19T21:38:21Z","cross_cats_sorted":[],"title_canon_sha256":"a3c7822ee5a7740aa0ca44248b4afa20b6ff81cbbf555954cb6ec5b1f6307853","abstract_canon_sha256":"0c5c02fd47ae9f6316f4671e973bc6bc15f87cd7259844b2efddca7cf7b350eb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T01:03:47.268304Z","signature_b64":"ALrMZvq/rAu9IctEKacVZBekK3l113h2lt5ohEuNxvRlfFDDe80eDdyXyJlvbUdlf5niyvqKnx9hiAQls9sXDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"390eb3e7053326c8284a3ce4fd88df10122f6b796dad200e21c6e3b1dec204a8","last_reissued_at":"2026-06-02T01:03:47.267480Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T01:03:47.267480Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FlowC2S: Flowing from Current to Succeeding Frames for Fast and Memory-Efficient Video Continuation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"FlowC2S flows directly from current video frames to succeeding ones, halving input size and outperforming prior methods with five evaluations.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Christian Sandor, Hovhannes Margaryan, Quentin Bammey","submitted_at":"2026-04-19T21:38:21Z","abstract_excerpt":"This paper introduces a novel methodology for generating fast and memory-efficient video continuations. Our method, dubbed FlowC2S, fine-tunes a pre-trained text-to-video flow model to learn a vector field between the current and succeeding video chunks. Two design choices are key. First, we introduce inherent optimal couplings, utilizing temporally adjacent video chunks during training as a practical proxy for true optimal couplings, resulting in straighter flows. Second, we incorporate target inversion, injecting the inverted latent of the target chunk into the input representation to streng"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"By flowing directly from current to succeeding frames, instead of the common combination of current frames with noise to generate a video continuation, we reduce the dimensionality of the model input by a factor of two. The proposed method, fine-tuned from LTXV and Wan, surpasses the state-of-the-art scores across quantitative evaluations with FID and FVD, with as few as five neural function evaluations.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That temporally adjacent video chunks serve as a practical proxy for true optimal couplings and produce straighter flows, and that injecting the inverted latent of the target chunk strengthens correspondences without introducing artifacts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"FlowC2S generates video continuations by flowing directly from current to next frames in a fine-tuned flow model, using adjacent chunks as optimal couplings and target inversion to cut input size in half and beat SOTA FID/FVD scores.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"FlowC2S flows directly from current video frames to succeeding ones, halving input size and outperforming prior methods with five evaluations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6e25533fec9cd210c41bd75c238b1b36661780afc1ee66269c85a3b515995870"},"source":{"id":"2604.17625","kind":"arxiv","version":2},"verdict":{"id":"494a1ece-0c80-4c7f-9119-7531c77c92ce","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T05:31:26.627692Z","strongest_claim":"By flowing directly from current to succeeding frames, instead of the common combination of current frames with noise to generate a video continuation, we reduce the dimensionality of the model input by a factor of two. The proposed method, fine-tuned from LTXV and Wan, surpasses the state-of-the-art scores across quantitative evaluations with FID and FVD, with as few as five neural function evaluations.","one_line_summary":"FlowC2S generates video continuations by flowing directly from current to next frames in a fine-tuned flow model, using adjacent chunks as optimal couplings and target inversion to cut input size in half and beat SOTA FID/FVD scores.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That temporally adjacent video chunks serve as a practical proxy for true optimal couplings and produce straighter flows, and that injecting the inverted latent of the target chunk strengthens correspondences without introducing artifacts.","pith_extraction_headline":"FlowC2S flows directly from current video frames to succeeding ones, halving input size and outperforming prior methods with five evaluations."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.17625/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"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":"2604.17625","created_at":"2026-06-02T01:03:47.267561+00:00"},{"alias_kind":"arxiv_version","alias_value":"2604.17625v2","created_at":"2026-06-02T01:03:47.267561+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.17625","created_at":"2026-06-02T01:03:47.267561+00:00"},{"alias_kind":"pith_short_12","alias_value":"HEHLHZYFGMTM","created_at":"2026-06-02T01:03:47.267561+00:00"},{"alias_kind":"pith_short_16","alias_value":"HEHLHZYFGMTMQKCK","created_at":"2026-06-02T01:03:47.267561+00:00"},{"alias_kind":"pith_short_8","alias_value":"HEHLHZYF","created_at":"2026-06-02T01:03:47.267561+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/HEHLHZYFGMTMQKCKHTSP3CG7CA","json":"https://pith.science/pith/HEHLHZYFGMTMQKCKHTSP3CG7CA.json","graph_json":"https://pith.science/api/pith-number/HEHLHZYFGMTMQKCKHTSP3CG7CA/graph.json","events_json":"https://pith.science/api/pith-number/HEHLHZYFGMTMQKCKHTSP3CG7CA/events.json","paper":"https://pith.science/paper/HEHLHZYF"},"agent_actions":{"view_html":"https://pith.science/pith/HEHLHZYFGMTMQKCKHTSP3CG7CA","download_json":"https://pith.science/pith/HEHLHZYFGMTMQKCKHTSP3CG7CA.json","view_paper":"https://pith.science/paper/HEHLHZYF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2604.17625&json=true","fetch_graph":"https://pith.science/api/pith-number/HEHLHZYFGMTMQKCKHTSP3CG7CA/graph.json","fetch_events":"https://pith.science/api/pith-number/HEHLHZYFGMTMQKCKHTSP3CG7CA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HEHLHZYFGMTMQKCKHTSP3CG7CA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HEHLHZYFGMTMQKCKHTSP3CG7CA/action/storage_attestation","attest_author":"https://pith.science/pith/HEHLHZYFGMTMQKCKHTSP3CG7CA/action/author_attestation","sign_citation":"https://pith.science/pith/HEHLHZYFGMTMQKCKHTSP3CG7CA/action/citation_signature","submit_replication":"https://pith.science/pith/HEHLHZYFGMTMQKCKHTSP3CG7CA/action/replication_record"}},"created_at":"2026-06-02T01:03:47.267561+00:00","updated_at":"2026-06-02T01:03:47.267561+00:00"}