SenseTime Open-Sources 8 Billion Parameter Multimodal Model

AI firm SenseTime has released its SenseNova U1.5 Lite model as open-source software. The 8-billion-parameter multimodal model integrates visual understanding with image generation and editing capabilities.

The model supports native 4K image output and is designed to process complex instructions involving subjects, counts, spatial relationships, text, layouts, and visual styles. This development occurs as competitors like Baidu and Alibaba also focus on releasing similar open-source AI solutions.

SenseTime states that the lightweight model enhances identity preservation and spatial structure control during image editing. It also introduces new controls, including bounding boxes, visual markers, and the use of multiple reference images.

SenseNova U1.5 Lite is available through GitHub, Hugging Face, and ModelScope. The release aims to promote transparency in AI development and foster research and application within the field.