{"paper":{"title":"WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"Text-to-image models struggle to apply world knowledge in generated images according to a dedicated new benchmark.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CV","authors_text":"Bin Lin, Bin Zhu, Chaoran Feng, Jiaqi Liao, Kunpeng Ning, Li Yuan, Mengren Zheng, Munan Ning, Peng Jin, Weiyang Jin, Yuwei Niu","submitted_at":"2025-03-10T12:47:53Z","abstract_excerpt":"Text-to-Image (T2I) models are capable of generating high-quality artistic creations and visual content. However, existing research and evaluation standards predominantly focus on image realism and shallow text-image alignment, lacking a comprehensive assessment of complex semantic understanding and world knowledge integration in text-to-image generation. To address this challenge, we propose \\textbf{WISE}, the first benchmark specifically designed for \\textbf{W}orld Knowledge-\\textbf{I}nformed \\textbf{S}emantic \\textbf{E}valuation. WISE moves beyond simple word-pixel mapping by challenging mo"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"our findings reveal significant limitations in their ability to effectively integrate and apply world knowledge during image generation, highlighting critical pathways for enhancing knowledge incorporation and application in next-generation T2I models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the 1000 crafted prompts and 25 subdomains provide unbiased, comprehensive tests of world knowledge integration without post-hoc selection effects or prompt engineering artifacts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Text-to-image models show significant limitations in integrating world knowledge, as measured by the new WISE benchmark and WiScore metric across 20 models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Text-to-image models struggle to apply world knowledge in generated images according to a dedicated new benchmark.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"edadc41a735de8c196995cd2dcdcbd3ff83b849ba526f3985ac59fcc8c849790"},"source":{"id":"2503.07265","kind":"arxiv","version":3},"verdict":{"id":"1adc6059-3f59-4f90-9467-c82a0595476a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T16:18:34.201092Z","strongest_claim":"our findings reveal significant limitations in their ability to effectively integrate and apply world knowledge during image generation, highlighting critical pathways for enhancing knowledge incorporation and application in next-generation T2I models.","one_line_summary":"Text-to-image models show significant limitations in integrating world knowledge, as measured by the new WISE benchmark and WiScore metric across 20 models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the 1000 crafted prompts and 25 subdomains provide unbiased, comprehensive tests of world knowledge integration without post-hoc selection effects or prompt engineering artifacts.","pith_extraction_headline":"Text-to-image models struggle to apply world knowledge in generated images according to a dedicated new benchmark."},"references":{"count":61,"sample":[{"doi":"","year":2023,"title":"Pixart-alpha: Fast training of diffusion transformer for photorealistic text-to-image synthe- sis, 2023","work_id":"6917650a-4a8f-4f92-88f8-efd7cd5388a6","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"BLIP3-o: A Family of Fully Open Unified Multimodal Models-Architecture, Training and Dataset","work_id":"86d896d2-592f-4d9b-938e-dfeb11f9388f","ref_index":2,"cited_arxiv_id":"2505.09568","is_internal_anchor":true},{"doi":"","year":2024,"title":"Next token prediction towards multimodal intelligence: A comprehensive survey","work_id":"5cdd7b0b-6c98-4848-901b-eab0a5c3f9e6","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Generative pretraining from pixels","work_id":"0747dcb4-e395-4187-9b0b-dadb91e6cb2b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling","work_id":"67d9e391-26d1-459e-ab56-07e60511c886","ref_index":5,"cited_arxiv_id":"2501.17811","is_internal_anchor":true}],"resolved_work":61,"snapshot_sha256":"577510af768e568d8e73cc1b69bd6894d00aa54f407977ae229de95ebe08860b","internal_anchors":23},"formal_canon":{"evidence_count":2,"snapshot_sha256":"94fb97ae272139d7986f88dd4fddf08517d5c0f8c287e222f9e2431bfb1bf164"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}