{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:YHOYE4ITP52HLAFCO5HLGKGWWS","short_pith_number":"pith:YHOYE4IT","schema_version":"1.0","canonical_sha256":"c1dd8271137f747580a2774eb328d6b4996c5cdd762cbb0ea9f0e444d510328e","source":{"kind":"arxiv","id":"2509.07295","version":4},"attestation_state":"computed","paper":{"title":"Reconstruction Alignment Improves Unified Multimodal Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Ji Xie, Luke Zettlemoyer, Trevor Darrell, Xudong Wang","submitted_at":"2025-09-08T23:59:32Z","abstract_excerpt":"Unified multimodal models (UMMs) unify visual understanding and generation within a single architecture. However, conventional training relies on image-text pairs (or sequences) whose captions are typically sparse and miss fine-grained visual details, even when they use hundreds of words to describe a simple image. We introduce Reconstruction Alignment (RECA), a resource-efficient post-training method that leverages visual understanding encoder embeddings as dense \"text prompts\", providing rich supervision without captions. Concretely, RECA conditions a UMM on its own visual understanding embe"},"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":"2509.07295","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-09-08T23:59:32Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"49749d33be401e02ee73903f8561295e28f3248fc2aed5c295183112ee387b37","abstract_canon_sha256":"656819747b01f13ea5bc3f487f14d8f7580548f9d4851da7005b0247aa600b0a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-26T01:15:14.369287Z","signature_b64":"4l/7JC6+TlK9judOjAbXf7xsNlrKhmPUIyMmO4Igzl/ckvmtS4zEXakKvcaIiSIFkXG55muH/9wK2tFZ/I3TDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c1dd8271137f747580a2774eb328d6b4996c5cdd762cbb0ea9f0e444d510328e","last_reissued_at":"2026-06-26T01:15:14.368840Z","signature_status":"signed_v1","first_computed_at":"2026-06-26T01:15:14.368840Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reconstruction Alignment Improves Unified Multimodal Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Ji Xie, Luke Zettlemoyer, Trevor Darrell, Xudong Wang","submitted_at":"2025-09-08T23:59:32Z","abstract_excerpt":"Unified multimodal models (UMMs) unify visual understanding and generation within a single architecture. However, conventional training relies on image-text pairs (or sequences) whose captions are typically sparse and miss fine-grained visual details, even when they use hundreds of words to describe a simple image. We introduce Reconstruction Alignment (RECA), a resource-efficient post-training method that leverages visual understanding encoder embeddings as dense \"text prompts\", providing rich supervision without captions. Concretely, RECA conditions a UMM on its own visual understanding embe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.07295","kind":"arxiv","version":4},"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/2509.07295/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":"2509.07295","created_at":"2026-06-26T01:15:14.368900+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.07295v4","created_at":"2026-06-26T01:15:14.368900+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.07295","created_at":"2026-06-26T01:15:14.368900+00:00"},{"alias_kind":"pith_short_12","alias_value":"YHOYE4ITP52H","created_at":"2026-06-26T01:15:14.368900+00:00"},{"alias_kind":"pith_short_16","alias_value":"YHOYE4ITP52HLAFC","created_at":"2026-06-26T01:15:14.368900+00:00"},{"alias_kind":"pith_short_8","alias_value":"YHOYE4IT","created_at":"2026-06-26T01:15:14.368900+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":14,"sample":[{"citing_arxiv_id":"2605.21487","citing_title":"Uni-Edit: Intelligent Editing Is A General Task For Unified Model Tuning","ref_index":40,"is_internal_anchor":true},{"citing_arxiv_id":"2605.21487","citing_title":"Uni-Edit: Intelligent Editing Is A General Task For Unified Model Tuning","ref_index":39,"is_internal_anchor":true},{"citing_arxiv_id":"2604.24763","citing_title":"Tuna-2: Pixel Embeddings Beat Vision Encoders for Multimodal Understanding and Generation","ref_index":45,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18714","citing_title":"Semantic Generative Tuning for Unified Multimodal Models","ref_index":77,"is_internal_anchor":true},{"citing_arxiv_id":"2511.22663","citing_title":"AIA: Rethinking Architecture Decoupling Strategy In Unified Multimodal Model","ref_index":45,"is_internal_anchor":true},{"citing_arxiv_id":"2601.08565","citing_title":"Rewriting Video: Text-Driven Reauthoring of Video Footage","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11400","citing_title":"UniPath: Adaptive Coordination of Understanding and Generation for Unified Multimodal Reasoning","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2604.24763","citing_title":"Tuna-2: Pixel Embeddings Beat Vision Encoders for Multimodal Understanding and Generation","ref_index":45,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05781","citing_title":"Steering Visual Generation in Unified Multimodal Models with Understanding Supervision","ref_index":64,"is_internal_anchor":true},{"citing_arxiv_id":"2604.19954","citing_title":"Camera Control for Text-to-Image Generation via Learning Viewpoint Tokens","ref_index":44,"is_internal_anchor":true},{"citing_arxiv_id":"2604.10949","citing_title":"Pseudo-Unification: Entropy Probing Reveals Divergent Information Patterns in Unified Multimodal Models","ref_index":69,"is_internal_anchor":true},{"citing_arxiv_id":"2604.06339","citing_title":"Evolution of Video Generative Foundations","ref_index":159,"is_internal_anchor":true},{"citing_arxiv_id":"2604.06870","citing_title":"RefineAnything: Multimodal Region-Specific Refinement for Perfect Local Details","ref_index":45,"is_internal_anchor":true},{"citing_arxiv_id":"2604.18168","citing_title":"Extending One-Step Image Generation from Class Labels to Text via Discriminative Text Representation","ref_index":30,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YHOYE4ITP52HLAFCO5HLGKGWWS","json":"https://pith.science/pith/YHOYE4ITP52HLAFCO5HLGKGWWS.json","graph_json":"https://pith.science/api/pith-number/YHOYE4ITP52HLAFCO5HLGKGWWS/graph.json","events_json":"https://pith.science/api/pith-number/YHOYE4ITP52HLAFCO5HLGKGWWS/events.json","paper":"https://pith.science/paper/YHOYE4IT"},"agent_actions":{"view_html":"https://pith.science/pith/YHOYE4ITP52HLAFCO5HLGKGWWS","download_json":"https://pith.science/pith/YHOYE4ITP52HLAFCO5HLGKGWWS.json","view_paper":"https://pith.science/paper/YHOYE4IT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.07295&json=true","fetch_graph":"https://pith.science/api/pith-number/YHOYE4ITP52HLAFCO5HLGKGWWS/graph.json","fetch_events":"https://pith.science/api/pith-number/YHOYE4ITP52HLAFCO5HLGKGWWS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YHOYE4ITP52HLAFCO5HLGKGWWS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YHOYE4ITP52HLAFCO5HLGKGWWS/action/storage_attestation","attest_author":"https://pith.science/pith/YHOYE4ITP52HLAFCO5HLGKGWWS/action/author_attestation","sign_citation":"https://pith.science/pith/YHOYE4ITP52HLAFCO5HLGKGWWS/action/citation_signature","submit_replication":"https://pith.science/pith/YHOYE4ITP52HLAFCO5HLGKGWWS/action/replication_record"}},"created_at":"2026-06-26T01:15:14.368900+00:00","updated_at":"2026-06-26T01:15:14.368900+00:00"}