{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:IDXTOZHB2CIV2VBSNRIYJY5ZFF","short_pith_number":"pith:IDXTOZHB","schema_version":"1.0","canonical_sha256":"40ef3764e1d0915d54326c5184e3b929707d8a3b0db70304a97667f5a1b278e8","source":{"kind":"arxiv","id":"2509.06942","version":3},"attestation_state":"computed","paper":{"title":"Directly Aligning the Full Diffusion Trajectory with Fine-Grained Human Preference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Chunyu Wang, Donghao Li, Qinglin Lu, Shiyi Zhang, Xiangwei Shen, Yansong Tang, Yingfang Zhang, Zhantao Yang, Zhimin Li","submitted_at":"2025-09-08T17:54:08Z","abstract_excerpt":"Recent studies have demonstrated the effectiveness of directly aligning diffusion models with human preferences using differentiable reward. However, they exhibit two primary challenges: (1) they rely on multistep denoising with gradient computation for reward scoring, which is computationally expensive, thus restricting optimization to only a few diffusion steps; (2) they often need continuous offline adaptation of reward models in order to achieve desired aesthetic quality, such as photorealism or precise lighting effects. To address the limitation of multistep denoising, we propose Direct-A"},"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":"2509.06942","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-09-08T17:54:08Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"9337f92092d9f30e3e9872cb3dc1b34568f42f618b07e64fcc8d886e54603437","abstract_canon_sha256":"90ce9d3a6a1b555a98c55f7ffea577f9fcaa13e4c4420af45837541d33fd847b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:09:24.391473Z","signature_b64":"azepqzlfcYMZ4gUIg0c+JQr46VltD0YYSvWsSJCjEd+mlwi/tTMURFQeZ9btS2e1Bhs7ZQnvfJuuyqSpqCSSCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"40ef3764e1d0915d54326c5184e3b929707d8a3b0db70304a97667f5a1b278e8","last_reissued_at":"2026-07-05T12:09:24.390959Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:09:24.390959Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Directly Aligning the Full Diffusion Trajectory with Fine-Grained Human Preference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Chunyu Wang, Donghao Li, Qinglin Lu, Shiyi Zhang, Xiangwei Shen, Yansong Tang, Yingfang Zhang, Zhantao Yang, Zhimin Li","submitted_at":"2025-09-08T17:54:08Z","abstract_excerpt":"Recent studies have demonstrated the effectiveness of directly aligning diffusion models with human preferences using differentiable reward. However, they exhibit two primary challenges: (1) they rely on multistep denoising with gradient computation for reward scoring, which is computationally expensive, thus restricting optimization to only a few diffusion steps; (2) they often need continuous offline adaptation of reward models in order to achieve desired aesthetic quality, such as photorealism or precise lighting effects. To address the limitation of multistep denoising, we propose Direct-A"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.06942","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2509.06942/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":"2509.06942","created_at":"2026-07-05T12:09:24.391013+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.06942v3","created_at":"2026-07-05T12:09:24.391013+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.06942","created_at":"2026-07-05T12:09:24.391013+00:00"},{"alias_kind":"pith_short_12","alias_value":"IDXTOZHB2CIV","created_at":"2026-07-05T12:09:24.391013+00:00"},{"alias_kind":"pith_short_16","alias_value":"IDXTOZHB2CIV2VBS","created_at":"2026-07-05T12:09:24.391013+00:00"},{"alias_kind":"pith_short_8","alias_value":"IDXTOZHB","created_at":"2026-07-05T12:09:24.391013+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.19839","citing_title":"When Preference Labels Fall Short: Aligning Diffusion Models from Real Data","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19839","citing_title":"When Preference Labels Fall Short: Aligning Diffusion Models from Real Data","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2509.23951","citing_title":"HunyuanImage 3.0 Technical Report","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15311","citing_title":"LeapAlign: Post-Training Flow Matching Models at Any Generation Step by Building Two-Step Trajectories","ref_index":43,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IDXTOZHB2CIV2VBSNRIYJY5ZFF","json":"https://pith.science/pith/IDXTOZHB2CIV2VBSNRIYJY5ZFF.json","graph_json":"https://pith.science/api/pith-number/IDXTOZHB2CIV2VBSNRIYJY5ZFF/graph.json","events_json":"https://pith.science/api/pith-number/IDXTOZHB2CIV2VBSNRIYJY5ZFF/events.json","paper":"https://pith.science/paper/IDXTOZHB"},"agent_actions":{"view_html":"https://pith.science/pith/IDXTOZHB2CIV2VBSNRIYJY5ZFF","download_json":"https://pith.science/pith/IDXTOZHB2CIV2VBSNRIYJY5ZFF.json","view_paper":"https://pith.science/paper/IDXTOZHB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.06942&json=true","fetch_graph":"https://pith.science/api/pith-number/IDXTOZHB2CIV2VBSNRIYJY5ZFF/graph.json","fetch_events":"https://pith.science/api/pith-number/IDXTOZHB2CIV2VBSNRIYJY5ZFF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IDXTOZHB2CIV2VBSNRIYJY5ZFF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IDXTOZHB2CIV2VBSNRIYJY5ZFF/action/storage_attestation","attest_author":"https://pith.science/pith/IDXTOZHB2CIV2VBSNRIYJY5ZFF/action/author_attestation","sign_citation":"https://pith.science/pith/IDXTOZHB2CIV2VBSNRIYJY5ZFF/action/citation_signature","submit_replication":"https://pith.science/pith/IDXTOZHB2CIV2VBSNRIYJY5ZFF/action/replication_record"}},"created_at":"2026-07-05T12:09:24.391013+00:00","updated_at":"2026-07-05T12:09:24.391013+00:00"}