cs.CV

SenseNova-U1.5: Towards Native Unified Visual Intelligence

Haiwen Diao, Jiahao Wang, Chenjing Ding, Hanming Deng, Jiangnan Chen, Ruixi Zhang, Ruohui Wang, Wenwen Tong, Xiangyu Fan, Yubo Wang, Yue Zhu, Yuwei Niu, Zhengqi Bai, Zhiqian Lin, Zhitao Yang, Zhongang Cai, Bo Yang, Chen Feng, Chengguang Lv, Guangjia Liu, Guanlin Wang, Hanyu Zhang, Haojia Yu, Hongcan Xiao, Hongli Wang, Huan Wu, Huaping Zhong, Jian Fang, Jianan Fan, Jiaqi Li, Jiefan Lu, Jing Zuo, Jingcheng Ni, Junxiang Xu, Linjun Dai, Mutian Xu, Peishen Yan, Penghao Wu, Ruijie Mao, Ruisi Wang, Shihao Bai, Shuang Yang, Shuya Yang, Shuyan Zheng, Silei Wu, Siying Li, Tao Chu, Tianbo Zhong, Tongxi Zhou, Weichao Luo, Weichen Fan, Wenhao Jia, Wenjie Gao, Xiangli Kong, Yan Li, Yang Yong, Zimo Wen, Zixuan Qian, Wenxiu Sun, Ruihao Gong, Quan Wang, Lewei Lu, Lei Yang, Ziwei Liu, Dahua Lin

arXiv: 2609.11929Published: September 10, 2026248 upvotes

View on arXiv Read PDF

AI summary

TL;DR

SenseNova-U1.5 is an 8B-parameter multimodal model designed for unified visual understanding and generation without relying on encoders or VAEs. It achieves high-fidelity image synthesis, accurate text rendering, and complex visual instruction following through native 4K resolution processing and multi-expert distillation.

Why It Matters

The model demonstrates that end-to-end unified architectures can handle diverse visual tasks with strong generalization, enabling more natural and precise visual creation and editing. This advances the development of AI systems capable of perceiving, reasoning, and generating visual content in a coherent, scalable way.

Key Findings

The model uses a native unified architecture without encoder or VAE components, enabling direct visual tokenization and generation.
It achieves high-quality image generation and editing at native 4K resolution, with strong preservation of subject identity and unmodified regions.
Multi-expert on-policy distillation improves performance across visual aesthetics, bilingual text rendering, infographic creation, and multi-reference editing.

Topics

multimodal

unified modeling

end-to-end

image generation

visual reasoning

Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.

SenseNova-U1.5: Towards Native Unified Visual Intelligence

arXiv: 2609.11929Published: September 10, 2026248 upvotes

TL;DR

Why It Matters

Key Findings

The model uses a native unified architecture without encoder or VAE components, enabling direct visual tokenization and generation.

It achieves high-quality image generation and editing at native 4K resolution, with strong preservation of subject identity and unmodified regions.

Multi-expert on-policy distillation improves performance across visual aesthetics, bilingual text rendering, infographic creation, and multi-reference editing.