{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:3H4R47TTTW5EWG6PPF7UEPMR4J","short_pith_number":"pith:3H4R47TT","schema_version":"1.0","canonical_sha256":"d9f91e7e739dba4b1bcf797f423d91e26b9f79705133b5705812091793909e9a","source":{"kind":"arxiv","id":"2505.21179","version":3},"attestation_state":"computed","paper":{"title":"Normalized Attention Guidance: Universal Negative Guidance for Diffusion Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Dar-Yen Chen, Hmrishav Bandyopadhyay, Kai Zou, Yi-Zhe Song","submitted_at":"2025-05-27T13:30:46Z","abstract_excerpt":"Negative guidance -- explicitly suppressing unwanted attributes -- remains a fundamental challenge in diffusion models, particularly in few-step sampling regimes. While Classifier-Free Guidance (CFG) works well in standard settings, it fails under aggressive sampling step compression due to divergent predictions between positive and negative branches. We present Normalized Attention Guidance (NAG), an efficient, training-free mechanism that applies extrapolation in attention space with L1-based normalization and refinement. NAG restores effective negative guidance where CFG collapses while mai"},"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":"2505.21179","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-05-27T13:30:46Z","cross_cats_sorted":[],"title_canon_sha256":"3166373539540c9c4e08da7e7ee447e0d5614948014d25c03f242d69a1e41590","abstract_canon_sha256":"baf43126cae1317bd79997a2e0e6da67d37339bba5a2a8d327b4ce3aed52bd34"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:14:29.567122Z","signature_b64":"xqugu7Yg6kfqQcstZLTFCvaHPCF3KKYe/vpLX3iLOKwLat640SoBaGZY5KAozEOQJDSQA8PQz6QTINR8OEWPDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d9f91e7e739dba4b1bcf797f423d91e26b9f79705133b5705812091793909e9a","last_reissued_at":"2026-07-05T11:14:29.566601Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:14:29.566601Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Normalized Attention Guidance: Universal Negative Guidance for Diffusion Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Dar-Yen Chen, Hmrishav Bandyopadhyay, Kai Zou, Yi-Zhe Song","submitted_at":"2025-05-27T13:30:46Z","abstract_excerpt":"Negative guidance -- explicitly suppressing unwanted attributes -- remains a fundamental challenge in diffusion models, particularly in few-step sampling regimes. While Classifier-Free Guidance (CFG) works well in standard settings, it fails under aggressive sampling step compression due to divergent predictions between positive and negative branches. We present Normalized Attention Guidance (NAG), an efficient, training-free mechanism that applies extrapolation in attention space with L1-based normalization and refinement. NAG restores effective negative guidance where CFG collapses while mai"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.21179","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/2505.21179/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":"2505.21179","created_at":"2026-07-05T11:14:29.566666+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.21179v3","created_at":"2026-07-05T11:14:29.566666+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.21179","created_at":"2026-07-05T11:14:29.566666+00:00"},{"alias_kind":"pith_short_12","alias_value":"3H4R47TTTW5E","created_at":"2026-07-05T11:14:29.566666+00:00"},{"alias_kind":"pith_short_16","alias_value":"3H4R47TTTW5EWG6P","created_at":"2026-07-05T11:14:29.566666+00:00"},{"alias_kind":"pith_short_8","alias_value":"3H4R47TT","created_at":"2026-07-05T11:14:29.566666+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.00647","citing_title":"Not All Prediction Targets Keep Training-Free Diffusion Guidance on the Manifold","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29390","citing_title":"Orthogonal Negative Guidance in Attention Feature Space for Text-to-Image Generation","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07026","citing_title":"Not all tokens contribute equally to diffusion learning","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3H4R47TTTW5EWG6PPF7UEPMR4J","json":"https://pith.science/pith/3H4R47TTTW5EWG6PPF7UEPMR4J.json","graph_json":"https://pith.science/api/pith-number/3H4R47TTTW5EWG6PPF7UEPMR4J/graph.json","events_json":"https://pith.science/api/pith-number/3H4R47TTTW5EWG6PPF7UEPMR4J/events.json","paper":"https://pith.science/paper/3H4R47TT"},"agent_actions":{"view_html":"https://pith.science/pith/3H4R47TTTW5EWG6PPF7UEPMR4J","download_json":"https://pith.science/pith/3H4R47TTTW5EWG6PPF7UEPMR4J.json","view_paper":"https://pith.science/paper/3H4R47TT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.21179&json=true","fetch_graph":"https://pith.science/api/pith-number/3H4R47TTTW5EWG6PPF7UEPMR4J/graph.json","fetch_events":"https://pith.science/api/pith-number/3H4R47TTTW5EWG6PPF7UEPMR4J/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3H4R47TTTW5EWG6PPF7UEPMR4J/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3H4R47TTTW5EWG6PPF7UEPMR4J/action/storage_attestation","attest_author":"https://pith.science/pith/3H4R47TTTW5EWG6PPF7UEPMR4J/action/author_attestation","sign_citation":"https://pith.science/pith/3H4R47TTTW5EWG6PPF7UEPMR4J/action/citation_signature","submit_replication":"https://pith.science/pith/3H4R47TTTW5EWG6PPF7UEPMR4J/action/replication_record"}},"created_at":"2026-07-05T11:14:29.566666+00:00","updated_at":"2026-07-05T11:14:29.566666+00:00"}