SenseNova U1.5-Lite-Preview – SenseTime's Open-Source Lightweight Multimodal Model Operational

SenseNova U1.5-Lite-Preview is a preview release of SenseTime's open-source lightweight native unified multimodal model. Built on the NEO-Unify architecture, it connects visual understanding, reasoning, generation, and editing with just 8B-MoT parameters. The model natively supports 4K image generation and delivers fine local textures, realistic real-world details, Chinese and English text generation, and complex layout capabilities. It is optimized for accurately following long natural-language and structured visual instructions, with the Prompt Enhance Skill further lowering the barrier to complex creation.

Native 4K generation: Supports end-to-end generation of ultra-high-resolution images, with details that remain clear when enlarged.
Fine visual details: Delivers more realistic material textures, lighting depth, and local details.
Chinese and English text generation: Accurately renders Chinese and English text and typography in complex layouts.
Image editing and iteration: Supports continuous iterative image editing based on natural-language instructions.
Prompt Enhance Skill: Automatically expands brief requirements into complete creative plans, lowering the barrier to writing instructions.

Follow the WeChat account and reply “开源” to join the AI open-source project discussion group.
NEO-Unify unified architecture: Jointly models language, visual semantics, and pixel generation in a single model, natively unifying understanding, reasoning, generation, and editing.
8B-MoT lightweight design: Uses a mixture-of-token architecture with just 8 billion parameters. It expands the boundaries of multimodal capabilities while remaining lightweight and demonstrates the continued scalability of a unified architecture.
Native pixel-language mapping: Trained end to end from pixels and language, the model learns a native mapping from language descriptions to visual structures.
Long-context visual control: Optimized to understand long natural-language and structured creative instructions, enabling precise execution of multi-constraint, hierarchical visual directions.

Get the model and documentation: Visit the GitHub repository for technical documentation and inference code, or download the 8B-MoT model weights and configuration files from Hugging Face.
Environment setup: Configure the runtime environment according to the documentation and load the model weights to initialize the model locally or in the cloud.
Basic image generation: Enter a natural-language description directly (Chinese and English are supported), and the model will natively generate a 4K image.
Complex creative instructions: Provide complete requirements covering the subject, layout, style, text, and constraints, and the model will generate the image according to the specified visual structure.
Prompt Enhance assistance: Invoke the experimental Prompt Enhance Skill to automatically expand a brief idea into a complete creative plan, then submit it to the model for execution.
Image editing: Upload an existing image and add natural-language editing instructions. The model performs continuous iterative editing while maintaining overall consistency.

Lightweight unification: Achieves integrated understanding, generation, and editing with just 8B-MoT parameters, reducing deployment costs.
4K ultra-high definition: Natively supports 4K resolution generation while maintaining detail and consistency at very large image sizes.
Strong instruction following: Reliably executes lengthy, complex visual instructions even without dedicated Prompt Enhance formatting training.
Realistic details: Significantly improves local textures, material lighting, and real-world detail rendering.
Text stability: Generates Chinese and English text and organizes complex layouts with greater accuracy and stability.

GitHub repository: https://github.com/OpenSenseNova/SenseNova-U1/blob/main/docs/u1.5_preview.md
Hugging Face model: https://huggingface.co/sensenova/SenseNova-U1.5-8B-MoT-Preview

Commercial brand visual design: Quickly generate posters, infographics, and brand marketing materials containing precise Chinese and English text, complex layouts, and specified visual styles to meet commercial creation needs requiring a high degree of control.
Ultra-wide narrative panoramas: Create ultra-wide continuous narrative content, such as historical timelines and cultural scrolls, in 4K ultra-high definition. Large amounts of text and icon details remain clear and stable when enlarged.
Architectural and product concept visualization: Produce highly realistic architectural exterior renderings, interior space concept images, and product showcase images with realistic material textures, lighting depth, and spatial atmosphere.
Photography-grade portraits and close-ups: Generate portrait photography, facial close-ups, and character-themed works with fine skin textures, natural lighting, and realistic details, while supporting complex compositions and specified styles.
Vintage and collage art: Create stylized visuals such as vintage scrapbook pages, collage posters, and natural-history illustrations featuring multiple integrated elements, accurately rendering details including paper textures, stamps, and handwritten text.