{"paper":{"title":"Accelerating Rectified Flow Models via Trajectory-Aware Caching","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"TACache accelerates rectified flow sampling by decomposing velocity changes into parallel and orthogonal parts to safely skip steps and reconstruct velocities from history.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Hongliang Lu, Kai Liu, Naiyang Guan, Renjing Pei, Xiao Liu, Yulun Zhang, Zhikai Chen, Zhixin Wang","submitted_at":"2026-05-16T03:44:58Z","abstract_excerpt":"Diffusion and rectified flow (RF) models generate high-fidelity images and videos, but their iterative velocity-field evaluations are computationally expensive. Existing caching methods accelerate sampling by skipping timesteps, yet their coarse approximations introduce accumulated errors over long skip intervals and degrade quality under aggressive acceleration. We propose TACache (Trajectory-Aware Cache), a training-free acceleration framework following a skip-then-compensate paradigm. TACache performs an orthogonal decomposition of discrete velocity acceleration along the RF trajectory into"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"TACache achieves up to 4.14 speedup on text-to-image generation and 2.11 speedup on text-to-video generation, with consistent improvements over prior cache-based methods on all reference-based fidelity metrics.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that offline cumulative variation thresholds on magnitude and direction indicators from the orthogonal decomposition can reliably bound skip intervals across diverse samples, and that combining these with a sample's historical orthogonal direction accurately reconstructs skipped velocities without model evaluations or accumulated error.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"TACache accelerates rectified flow sampling up to 4.14x for text-to-image and 2.11x for text-to-video via offline skip scheduling from cumulative variation thresholds and online velocity reconstruction using historical orthogonal directions.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"TACache accelerates rectified flow sampling by decomposing velocity changes into parallel and orthogonal parts to safely skip steps and reconstruct velocities from history.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"82d28c95b1225380864635e0f94aaa70aae3258a5c9039373bbfa0e5669ce57a"},"source":{"id":"2605.16789","kind":"arxiv","version":1},"verdict":{"id":"41a47156-e9ce-4916-ae6d-4871352bf010","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T21:27:56.083008Z","strongest_claim":"TACache achieves up to 4.14 speedup on text-to-image generation and 2.11 speedup on text-to-video generation, with consistent improvements over prior cache-based methods on all reference-based fidelity metrics.","one_line_summary":"TACache accelerates rectified flow sampling up to 4.14x for text-to-image and 2.11x for text-to-video via offline skip scheduling from cumulative variation thresholds and online velocity reconstruction using historical orthogonal directions.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that offline cumulative variation thresholds on magnitude and direction indicators from the orthogonal decomposition can reliably bound skip intervals across diverse samples, and that combining these with a sample's historical orthogonal direction accurately reconstructs skipped velocities without model evaluations or accumulated error.","pith_extraction_headline":"TACache accelerates rectified flow sampling by decomposing velocity changes into parallel and orthogonal parts to safely skip steps and reconstruct velocities from history."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16789/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T22:01:19.626600Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T21:40:53.267236Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T19:01:56.296349Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.432046Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"6781f852ae41dfc00ccb35dabb734d96e72632a792406643167c637ccb8733cc"},"references":{"count":28,"sample":[{"doi":"","year":2025,"title":"FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space","work_id":"5dfe19d5-3541-4803-8fe9-3c8b9e29b281","ref_index":1,"cited_arxiv_id":"2506.15742","is_internal_anchor":true},{"doi":"","year":2026,"title":"DiCache: Let diffusion model determine its own cache","work_id":"dc9adac5-5de3-410e-8232-894fad127fb0","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Emerging Properties in Unified Multimodal Pretraining","work_id":"e0cfd82c-f5d4-44fd-b531-ec73ab0a805b","ref_index":3,"cited_arxiv_id":"2505.14683","is_internal_anchor":true},{"doi":"","year":2025,"title":"Acceleration-aware sampling for few-step rectified flow models, 2025","work_id":"12eb4b11-9413-4a58-a976-b5edf3f11d46","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"GenEval: An object-focused framework for evaluating text-to-image alignment.NeurIPS, 2023","work_id":"eb43d79e-2b67-47eb-9e1d-7d72e74206eb","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":28,"snapshot_sha256":"a59a92dfa9f4d1c7fcf43bbfcf36db61bde7b2445528de7dc784c649c283fcf6","internal_anchors":7},"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"}