商汤预览SenseNova-U1 Pro：原生8K输出，跨5种场景直接对标GPT-Image-2
英文摘要
SenseTime previewed its next-gen multimodal model SenseNova-U1 Pro, claiming native 8K resolution output versus GPT-Image-2's 4K. The model uses a unified 'Understand-Generate-Action' architecture targeting professional design workflows. Direct comparisons showed U1 Pro outperforming GPT-Image-2 in an infographic, a scroll painting layout, a magazine spread, an academic poster, and a high-resolution storyboard. The model also generated the entire 20+ page shareholder meeting presentation end-to-end. Invite testing is slated to begin in July 2026.
中文摘要
商汤预览了下一代多模态模型SenseNova-U1 Pro，宣称支持原生8K分辨率输出（对比GPT-Image-2的原生4K）。该模型采用统一的“理解-生成-行动”架构，瞄准专业设计工作流。直接对比显示U1 Pro在信息图、长卷画布局、杂志跨页、学术海报和高分辨率分镜故事板共5个场景中优于GPT-Image-2。该模型还端到端生成了整场股东会20余页的PPT。邀约测试将于2026年7月启动。
关键要点
Native 8K resolution output, double the claimed 4K native of GPT-Image-2.
原生8K分辨率输出，是GPT-Image-2声称的原生4K的两倍。
Unified 'Understand-Generate-Action' multimodal architecture for professional design tasks.
统一的“理解-生成-行动”多模态架构，用于专业设计任务。
Direct comparison across 5 scenarios: infographic, Chinese painting scroll, magazine spread, academic poster, and storyboard, with U1 Pro showing higher fidelity and detail retention.
在信息图、国画卷轴、杂志跨页、学术海报和故事板5个场景中直接对比，U1 Pro展现了更高的保真度和细节保留能力。
Generated the entire 20+ page shareholder meeting PPT deck end-to-end, including planning, layout, and number accuracy.
端到端生成了整场股东会20多页的PPT，包括策划、版式和数字准确性。
Invite testing begins July 2026.
2026年7月启动邀约测试。