UniVerse: Empower Unified Generation with Reasoning and Knowledge

Current text-to-image (T2I) generation models often struggle with prompts that require complex reasoning or specialized knowledge, failing to accurately interpret implicit user intent. To bridge this gap, we introduce UniVerse, a largescale dataset designed to empower T2I generation in unified multimodal models (UMMs) with reasoning and knowledge. The dataset contains 120k pairs of text triplet and image. The text triplet consists of (1) an implicit prompt, which requires reasoning or knowledge to decipher its underlying meaning; (2) a reasoning chain, which provides a step-by-step analysis to resolve the implicit prompt; and (3)

an explicit prompt, a clear and straightforward visual description prepared for T2I generation. UniVerse is meticulously constructed: 65k samples are dedicated to reasoning, specifically targeting arithmetic reasoning, spatial-attribute relationship reasoning, deductive reasoning (cause to effect), and abductive reasoning (effect to cause); and 55k samples are focused on specialized knowledge, including scientific disciplines, spatial-temporal concepts, and entity knowledge. To validate the effectiveness of our dataset, we finetune Bagel [6], a unified multimodal model, on Uni-Verse. Results from multiple benchmarks show significant and consistent improvements in both reasoning-oriented and knowledge-oriented generation. These findings suggest that training with implicit prompts and intermediate reasoning chains are key steps toward developing more intel-This CVPR paper is the Open Access version, provided by the Computer Vision Foundation.

Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore.