{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:WA3SL523UHMPNCX3AA2IGEG5Q4","short_pith_number":"pith:WA3SL523","schema_version":"1.0","canonical_sha256":"b03725f75ba1d8f68afb00348310dd870075e9c8e247dc102acdb515ba124484","source":{"kind":"arxiv","id":"2205.02655","version":2},"attestation_state":"computed","paper":{"title":"Language Models Can See: Plugging Visual Controls in Text Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Dani Yogatama, Fangyu Liu, Lingpeng Kong, Nigel Collier, Tian Lan, Yahui Liu, Yan Wang, Yixuan Su","submitted_at":"2022-05-05T13:56:18Z","abstract_excerpt":"Generative language models (LMs) such as GPT-2/3 can be prompted to generate text with remarkable quality. While they are designed for text-prompted generation, it remains an open question how the generation process could be guided by modalities beyond text such as images. In this work, we propose a training-free framework, called MAGIC (iMAge-Guided text generatIon with CLIP), for plugging in visual controls in the generation process and enabling LMs to perform multimodal tasks (e.g., image captioning) in a zero-shot manner. MAGIC is a simple yet efficient plug-and-play framework, which direc"},"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":"2205.02655","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2022-05-05T13:56:18Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"0f5f2106a480e0869f41621abeaefcfddb203e30e3af38d65e3f3e90daab722e","abstract_canon_sha256":"35adc59ad9d9b7a6464dbf2e7e44551fd52813d0ec82d8e5556ffccf3234f4a9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:27:31.135227Z","signature_b64":"pess32UseUlsu+UaDVupfGACV2T6awFbj8d8ir23ZMvAf8JbNq08LmxQFdPzDQYdbt0vYlefmWRyEfCCwahvDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b03725f75ba1d8f68afb00348310dd870075e9c8e247dc102acdb515ba124484","last_reissued_at":"2026-07-05T04:27:31.134748Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:27:31.134748Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Language Models Can See: Plugging Visual Controls in Text Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Dani Yogatama, Fangyu Liu, Lingpeng Kong, Nigel Collier, Tian Lan, Yahui Liu, Yan Wang, Yixuan Su","submitted_at":"2022-05-05T13:56:18Z","abstract_excerpt":"Generative language models (LMs) such as GPT-2/3 can be prompted to generate text with remarkable quality. While they are designed for text-prompted generation, it remains an open question how the generation process could be guided by modalities beyond text such as images. In this work, we propose a training-free framework, called MAGIC (iMAge-Guided text generatIon with CLIP), for plugging in visual controls in the generation process and enabling LMs to perform multimodal tasks (e.g., image captioning) in a zero-shot manner. MAGIC is a simple yet efficient plug-and-play framework, which direc"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2205.02655","kind":"arxiv","version":2},"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/2205.02655/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":"2205.02655","created_at":"2026-07-05T04:27:31.134799+00:00"},{"alias_kind":"arxiv_version","alias_value":"2205.02655v2","created_at":"2026-07-05T04:27:31.134799+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2205.02655","created_at":"2026-07-05T04:27:31.134799+00:00"},{"alias_kind":"pith_short_12","alias_value":"WA3SL523UHMP","created_at":"2026-07-05T04:27:31.134799+00:00"},{"alias_kind":"pith_short_16","alias_value":"WA3SL523UHMPNCX3","created_at":"2026-07-05T04:27:31.134799+00:00"},{"alias_kind":"pith_short_8","alias_value":"WA3SL523","created_at":"2026-07-05T04:27:31.134799+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2511.20657","citing_title":"Intelligent Agents with Emotional Intelligence: Current Trends, Challenges, and Future Prospects","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2204.00598","citing_title":"Socratic Models: Composing Zero-Shot Multimodal Reasoning with Language","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2305.16355","citing_title":"PandaGPT: One Model To Instruction-Follow Them All","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2602.07026","citing_title":"Modality Gap-Driven Subspace Alignment Training Paradigm For Multimodal Large Language Models","ref_index":22,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WA3SL523UHMPNCX3AA2IGEG5Q4","json":"https://pith.science/pith/WA3SL523UHMPNCX3AA2IGEG5Q4.json","graph_json":"https://pith.science/api/pith-number/WA3SL523UHMPNCX3AA2IGEG5Q4/graph.json","events_json":"https://pith.science/api/pith-number/WA3SL523UHMPNCX3AA2IGEG5Q4/events.json","paper":"https://pith.science/paper/WA3SL523"},"agent_actions":{"view_html":"https://pith.science/pith/WA3SL523UHMPNCX3AA2IGEG5Q4","download_json":"https://pith.science/pith/WA3SL523UHMPNCX3AA2IGEG5Q4.json","view_paper":"https://pith.science/paper/WA3SL523","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2205.02655&json=true","fetch_graph":"https://pith.science/api/pith-number/WA3SL523UHMPNCX3AA2IGEG5Q4/graph.json","fetch_events":"https://pith.science/api/pith-number/WA3SL523UHMPNCX3AA2IGEG5Q4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WA3SL523UHMPNCX3AA2IGEG5Q4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WA3SL523UHMPNCX3AA2IGEG5Q4/action/storage_attestation","attest_author":"https://pith.science/pith/WA3SL523UHMPNCX3AA2IGEG5Q4/action/author_attestation","sign_citation":"https://pith.science/pith/WA3SL523UHMPNCX3AA2IGEG5Q4/action/citation_signature","submit_replication":"https://pith.science/pith/WA3SL523UHMPNCX3AA2IGEG5Q4/action/replication_record"}},"created_at":"2026-07-05T04:27:31.134799+00:00","updated_at":"2026-07-05T04:27:31.134799+00:00"}