{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:P767SZHFHSZVSPIGAYPXV4PWEC","short_pith_number":"pith:P767SZHF","schema_version":"1.0","canonical_sha256":"7ffdf964e53cb3593d06061f7af1f6208fd49a73b4abcc5d3d1f4b5b19cbde13","source":{"kind":"arxiv","id":"2504.16080","version":1},"attestation_state":"computed","paper":{"title":"From Reflection to Perfection: Scaling Inference-Time Optimization for Text-to-Image Diffusion Models via Reflection Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Hongsheng Li, Le Zhuo, Liangbing Zhao, Mohamed Elhoseiny, Peng Gao, Renrui Zhang, Sayak Paul, Yi Xin, Yue Liao","submitted_at":"2025-04-22T17:58:07Z","abstract_excerpt":"Recent text-to-image diffusion models achieve impressive visual quality through extensive scaling of training data and model parameters, yet they often struggle with complex scenes and fine-grained details. Inspired by the self-reflection capabilities emergent in large language models, we propose ReflectionFlow, an inference-time framework enabling diffusion models to iteratively reflect upon and refine their outputs. ReflectionFlow introduces three complementary inference-time scaling axes: (1) noise-level scaling to optimize latent initialization; (2) prompt-level scaling for precise semanti"},"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":"2504.16080","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-04-22T17:58:07Z","cross_cats_sorted":[],"title_canon_sha256":"c6a1b55467ac5cb9bd507b7b39ab0365697170902e4366120e01e2c4f73b6d49","abstract_canon_sha256":"b2769d5dab2aaebef72d6269f69ad6b825f9e27167ea874a41f11e6d536f726c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:52:35.204753Z","signature_b64":"cvYPvAtdMPKLKOgYzLbIVxG7xvWj0QG5kv2Z5KSzVtIN7wjSRSxpiV3XgAYRJgG5iYNBtkv9/HlMny764Uy7AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7ffdf964e53cb3593d06061f7af1f6208fd49a73b4abcc5d3d1f4b5b19cbde13","last_reissued_at":"2026-07-05T10:52:35.204283Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:52:35.204283Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"From Reflection to Perfection: Scaling Inference-Time Optimization for Text-to-Image Diffusion Models via Reflection Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Hongsheng Li, Le Zhuo, Liangbing Zhao, Mohamed Elhoseiny, Peng Gao, Renrui Zhang, Sayak Paul, Yi Xin, Yue Liao","submitted_at":"2025-04-22T17:58:07Z","abstract_excerpt":"Recent text-to-image diffusion models achieve impressive visual quality through extensive scaling of training data and model parameters, yet they often struggle with complex scenes and fine-grained details. Inspired by the self-reflection capabilities emergent in large language models, we propose ReflectionFlow, an inference-time framework enabling diffusion models to iteratively reflect upon and refine their outputs. ReflectionFlow introduces three complementary inference-time scaling axes: (1) noise-level scaling to optimize latent initialization; (2) prompt-level scaling for precise semanti"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.16080","kind":"arxiv","version":1},"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/2504.16080/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":"2504.16080","created_at":"2026-07-05T10:52:35.204341+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.16080v1","created_at":"2026-07-05T10:52:35.204341+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.16080","created_at":"2026-07-05T10:52:35.204341+00:00"},{"alias_kind":"pith_short_12","alias_value":"P767SZHFHSZV","created_at":"2026-07-05T10:52:35.204341+00:00"},{"alias_kind":"pith_short_16","alias_value":"P767SZHFHSZVSPIG","created_at":"2026-07-05T10:52:35.204341+00:00"},{"alias_kind":"pith_short_8","alias_value":"P767SZHF","created_at":"2026-07-05T10:52:35.204341+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17888","citing_title":"MathVis-Fine: Aligning Visual Supervision with Necessity via Progressive Dependency-Guided Training for Multimodal Mathematical Reasoning","ref_index":104,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14876","citing_title":"Unlocking Complex Visual Generation via Closed-Loop Verified Reasoning","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2506.18871","citing_title":"OmniGen2: Towards Instruction-Aligned Multimodal Generation","ref_index":95,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/P767SZHFHSZVSPIGAYPXV4PWEC","json":"https://pith.science/pith/P767SZHFHSZVSPIGAYPXV4PWEC.json","graph_json":"https://pith.science/api/pith-number/P767SZHFHSZVSPIGAYPXV4PWEC/graph.json","events_json":"https://pith.science/api/pith-number/P767SZHFHSZVSPIGAYPXV4PWEC/events.json","paper":"https://pith.science/paper/P767SZHF"},"agent_actions":{"view_html":"https://pith.science/pith/P767SZHFHSZVSPIGAYPXV4PWEC","download_json":"https://pith.science/pith/P767SZHFHSZVSPIGAYPXV4PWEC.json","view_paper":"https://pith.science/paper/P767SZHF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.16080&json=true","fetch_graph":"https://pith.science/api/pith-number/P767SZHFHSZVSPIGAYPXV4PWEC/graph.json","fetch_events":"https://pith.science/api/pith-number/P767SZHFHSZVSPIGAYPXV4PWEC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/P767SZHFHSZVSPIGAYPXV4PWEC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/P767SZHFHSZVSPIGAYPXV4PWEC/action/storage_attestation","attest_author":"https://pith.science/pith/P767SZHFHSZVSPIGAYPXV4PWEC/action/author_attestation","sign_citation":"https://pith.science/pith/P767SZHFHSZVSPIGAYPXV4PWEC/action/citation_signature","submit_replication":"https://pith.science/pith/P767SZHFHSZVSPIGAYPXV4PWEC/action/replication_record"}},"created_at":"2026-07-05T10:52:35.204341+00:00","updated_at":"2026-07-05T10:52:35.204341+00:00"}