SenseNova U1.5-Lite-Preview: SenseTime открыла мультимодальный ИИ с нативной генерацией изображений

SenseTime (商汤科技) on August 3, 2026 open-sourced SenseNova U1.5-Lite-Preview — a preview version of a lightweight natively-unified multimodal model based on the NEO-Unify architecture. This is a continuation of the April release of SenseNova U1 with enhanced image generation and editing capabilities.

What Is SenseNova U1 and the NEO-Unify Architecture

In April 2026, SenseTime first fully revealed the natively-unified multimodal approach based on the NEO-Unify architecture. The key idea: instead of a chain of specialized models — a single neural network jointly trained on three types of data simultaneously.

• April 2026 — release of SenseNova U1 with the first complete implementation of NEO-Unify
• A single neural network combines linguistic semantics, visual semantics, and pixel generation
• The model natively performs visual understanding, reasoning, and image generation
• August 3, 2026 — open source release of the lightweight version U1.5-Lite-Preview

Joint training across three modalities allows the model to maintain a unified representation of context. According to the developers' concept, this provides an advantage in complex tasks where image understanding directly affects generation quality.

What Changed in U1.5 Compared to the Original

SenseNova U1.5-Lite-Preview advances the April U1 in two key directions. According to SenseTime, in the months following the launch, the team focused specifically on improving image generation and editing — these components underwent the most significant changes.

"Over the past months, we have consistently strengthened the model's

capabilities for image generation and editing and are now opening a preview version of the lightweight model to the community," — stated in the official SenseTime release on the Jiqizhixin platform.

The "Lite" suffix indicates a lightweight variant of the architecture, designed for a broader audience of researchers and developers who do not have powerful infrastructure for flagship models. The "Preview" status means this is a preliminary release — the final version is yet to come.

Why Native Unification Is Better Than a Pipeline

Most multimodal systems still work as a pipeline: an image understanding block converts a picture into text, a language model processes it, a generative block renders the result. At each junction, context is lost and errors accumulate.

The natively-unified NEO-Unify architecture eliminates these junctions: the model sees the task as a whole — text, visual context, and pixel result — and processes them jointly. This allows image editing that accounts for subtle visual context that would be lost during conversion to text in a pipeline approach.

What This Means

The open-source release of SenseNova U1.5-Lite-Preview makes natively-unified multimodal models available for direct research. Developers can study SenseTime's architectural solution, fine-tune the model on their own data, or use it as a base for experiments. For SenseTime, this is a way to obtain community feedback before the final commercial release.

Frequently Asked Questions

What Is Natively-Unified Multimodality?

This is an approach in which a single neural network is trained to simultaneously work with text, visual data, and pixel generation without intermediate pipelines. SenseTime implemented this principle in the NEO-Unify architecture, starting from April 2026 in the SenseNova U1 model.

How Does U1.5-Lite-Preview Differ from SenseNova U1?

U1.5-Lite-Preview is a lightweight version with improved image generation and editing capabilities developed over the months following the April launch. "Lite" means a reduced model size for a wider audience; "Preview" means this is an open preliminary release for the research community, not a final commercial product.