SenseTime Open-Sourced SenseNova U1.5 Lite, an 8B Multimodal Model With Native 4K Image Output

TechNode·medium signalXBlueskyLinkedInCopy link

SenseTime released SenseNova U1.5 Lite on 2026-08-21, an 8-billion-parameter model combining visual understanding, image generation and editing in one system with native 4K output. It is built to respect constraints on subjects, counts, spatial relationships, text, layouts and visual styles, and adds control via bounding boxes, visual markers and multiple reference images, with claimed improvements to identity preservation and spatial structure during edits. Weights are on GitHub, Hugging Face and ModelScope, which puts a 4K-capable edit-and-generate model in reach of a single consumer GPU.