arXiv:2511.18271v3 Announce Type: replace 
Abstract: Text-to-image (T2I) models today are capable of producing photorealistic, instruction-following images, yet they still frequently fail on prompts that require implicit world knowledge. Existing evaluation protocols either emphasize compositional alignment or rely on single-round VQA-based scoring, leaving critical dimensions such as knowledge grounding, multi-physics interactions, and auditable evidence-substantially undertested. To address these limitations, we introduce PicWorld, the first comprehensive benchmark that assesses the grasp of implicit world knowledge and physical causal reasoning of T2I models. This benchmark consists of 1,100 prompts across three core categories. To facilitate fine-grained evaluation, we propose PW-Agent, an evidence-grounded multi-agent evaluator to hierarchically assess images on their physical realism and logical consistency by decomposing prompts into verifiable visual evidence. We conduct a thorough analysis of 17 mainstream T2I models on PicWorld, illustrating that they universally exhibit a fundamental limitation in their capacity for implicit world knowledge and physical causal reasoning to varying degrees. The findings highlight the need for reasoning-aware, knowledge-integrative architectures in future T2I systems.

تم تقديم معيار جديد يسمى PicWorld لتقييم قدرات المعرفة الضمنية بالعالم والتفكير الفيزيائي لنماذج النص إلى صورة (T2I). يتضمن هذا المعيار 1,100 طلبًا مصنفة إلى ثلاث فئات رئيسية، بهدف معالجة القيود التي تعاني منها بروتوكولات التقييم الحالية التي غالبًا ما تتجاهل أبعادًا حاسمة مثل تأصيل المعرفة والتفاعلات متعددة الفيزياء.

Se ha introducido un nuevo benchmark llamado PicWorld para evaluar las capacidades de conocimiento implícito del mundo y razonamiento físico de los modelos de texto a imagen (T2I). Este benchmark incluye 1,100 prompts clasificados en tres categorías principales, con el objetivo de abordar las limitaciones de los protocolos de evaluación existentes que a menudo pasan por alto dimensiones críticas como la fundamentación del conocimiento y las interacciones multi-físicas.

Un nouveau benchmark appelé PicWorld a été introduit pour évaluer les capacités de connaissance implicite du monde et de raisonnement physique des modèles de texte à image (T2I). Ce benchmark comprend 1 100 prompts classés en trois catégories principales, visant à remédier aux limitations des protocoles d'évaluation existants qui négligent souvent des dimensions critiques telles que l'ancrage des connaissances et les interactions multi-physiques.

A new benchmark called PicWorld has been introduced to evaluate the implicit world knowledge and physical reasoning capabilities of text-to-image (T2I) models. This benchmark includes 1,100 prompts categorized into three core areas, aiming to address the limitations of existing evaluation protocols that often overlook critical dimensions such as knowledge grounding and multi-physics interactions.

Beyond Words and Pixels: A Benchmark for Implicit World Knowledge Reasoning in Generative Models

One More Thing in AI – Your Shortcut to AI Mastery

Beyond Words and Pixels: A Benchmark for Implicit World Knowledge Reasoning in Generative Models

Was this article worth reading? Share it

One More Thing in AI

LucidQuery AI

4o Image Gen

AIPortalX

GPTHumanizer

Republiclabs.ai

Ready to build your own newsroom?