AI News Feed
Market watch
LG UltraFine Evo 32U990A-S Hits Record-Low $1,099 on Amazon, TechRadar ReportsNewegg Labor Day Deal: 32GB Kingston Fury Beast DDR4-3200 Kit Nearly Half Price at $179.99 with CodeGalaxy Z Fold 8 inner screen corners reported with excessive giveHuawei unveils Mate XT 2 tri-fold with privacy-screen edition as Apple prepares folding iPhoneWall Street Analysts Recommend Nvidia, Uber and Marvell for Long-Term GrowthLenovo, AMD and Huawei Put Local Large-Model AI at Center of New Devices
LG UltraFine Evo 32U990A-S Hits Record-Low $1,099 on Amazon, TechRadar ReportsNewegg Labor Day Deal: 32GB Kingston Fury Beast DDR4-3200 Kit Nearly Half Price at $179.99 with CodeGalaxy Z Fold 8 inner screen corners reported with excessive giveHuawei unveils Mate XT 2 tri-fold with privacy-screen edition as Apple prepares folding iPhoneWall Street Analysts Recommend Nvidia, Uber and Marvell for Long-Term GrowthLenovo, AMD and Huawei Put Local Large-Model AI at Center of New Devices
Large Language Models

SenseTime Open-Sources SenseNova U1.5 Lite with Ultra-Long Instruction Support and Native 4K

SenseTime has open-sourced SenseNova U1.5 Lite, a lightweight multimodal model supporting 3-4k character instructions and native 4K output for stable visual creation.

AI News Feed · 2026-08-22 05:52

The model natively supports 3-4k character context length, breaking through the bottleneck where most open-source models tend to crash or miss details when instructions exceed 1k characters. It can simultaneously handle multiple constraints such as subject, quantity, spatial relationships, text, layout, and style, improving the execution stability of complex visual tasks. It also delivers higher-quality generation with better composition, color, material, light and shadow, realism, and local details.

SenseNova U1.5 Lite enhances native image editing by preserving subject identity, spatial structure, layout relationships, and non-edited areas, while improving local modification, element replacement, text refinement, and multi-reference image editing. It strengthens Chinese and English text rendering, posters, infographics, brand visuals, and multi-text layout, moving from content generation to complete visual expression.

The model supports Bounding Box, Visual Marker, and single-image or multi-image references, allowing more accurate editing of specified regions and objects. It also enables native 4K high-resolution output, balancing overall composition with extremely fine textures, small text, and light refraction.

In practical tests, the model demonstrated strong performance in creative poster generation, high-density infographic creation, and artistic detail reproduction. For example, it can transform a sketch into a retro comic illustration while strictly following a long prompt, or redesign a poster with a new theme while preserving the original layout. It also supports multi-image fusion, such as combining a cat and a motorcycle into a coherent scene, and editing infographic elements like replacing band members while keeping the original structure.

The model is an 8B-parameter lightweight architecture. SenseTime stated that it surpasses models of the same scale in instruction following and image editing consistency, and excels in text rendering and complex layout. The model is now globally open-sourced and available on GitHub, Hugging Face, ModelScope, and SenseNova Studio for developers and creators to deploy and experience.