SenseNova U1.5-Lite-Preview by SenseTime: 8B Model Supports Native 4K Image Generation

SenseTime has announced the open-source preview of its lightweight, unified multimodal model SenseNova U1.5-Lite-Preview. As a systematic iteration of SenseNova U1, this model is based on the NEO-unify architecture, achieving native 4K image generation, more detailed texture expression, more accurate Chinese and English text generation, and more stable image editing capabilities with only an 8B-MoT parameter scale.

According to the introduction, the SenseNova U1.5-Lite-Preview significantly improves the model's ability to understand long natural language descriptions and structured visual instructions, enabling more precise visual control according to complex requirements. In terms of image editing, the new model supports style transfer from reference images, combined editing with multiple reference images, and interactive precise adjustment, allowing AI image creation to shift from one-time generation to continuous iteration and optimization.

Compared to the previous generation model, U1.5-Lite-Preview has achieved improvements in multiple evaluation benchmarks. SenseTime stated that this model can balance generation quality and interaction capabilities with a smaller model size, further lowering the entry barrier for the application of multimodal AI technology.

Currently, SenseNova U1.5-Lite-Preview is available for download on GitHub, Hugging Face, and ModelScope community, allowing developers and researchers to perform secondary development based on the model. At the same time, the delivery-level creative model for professional creators and enterprise users SenseNova U1Pro is in a tight invitation testing phase.

As the competition in multimodal models gradually shifts from parameter scale to efficiency, control capability, and practical application experience, SenseTime's open-source of this lightweight model also reflects the industry's trend of promoting high-performance AI capabilities toward a broader developer ecosystem.

SenseNova-Vision by SenseTime Open-Sources a Unified Vision Large Model, Achieving Outstanding Performance in Four Core Visual Tasks with a Single Model

SenseTime has released and open-sourced the "SenseNova-Vision" vision large model, whose core lies in natively integrating vision capabilities into a general foundation model, breaking the fragmented mode of packaging expert models such as detection and segmentation. This model achieves dominant performance across multiple evaluation tasks in four fields, marking a critical upgrade in visual tasks toward a unified native architecture.

16GB Memory, Local Instant Response! Google Releases Gemma 4 12B Revolutionary Encoder-Free Architecture Ignites Open Source Community

Google releases a new multimodal model Gemma 4 12B, revolutionizing the traditional architecture by eliminating the separate encoder component, achieving efficient local deployment and inference on consumer-level hardware. This breakthrough significantly reduces the computational complexity of multimodal models, improves processing speed, and marks a new stage in the open source large model ecosystem.

New Developments in Domestic Large Models: MiniMax Launches the '10x Team' Program to Reward Global Top Experts

MiniMax (Xiyu Technology) has launched the '10x Team' global talent collaboration program, aiming to gather top experts from various industries, combine industry expertise with cutting-edge AI technology, and promote the application of large models in vertical fields, extending productivity from general to specialized scenarios, achieving a tenfold increase in industry efficiency. It also opens up multimodal core resources to verify the value of industry insights.

NVIDIA Releases the Nemotron 3 Series Open-Source Models: Inference Efficiency Surges 5 Times

NVIDIA announced a significant expansion of its open-source model family at the 2026 GTC conference, with the key release of the Nemotron 3 series multimodal models. Among them, Nemotron 3 Ultra is optimized based on the Blackwell architecture, achieving a fivefold improvement in throughput efficiency, specifically designed for complex code assistance and enterprise workflows. Meanwhile, the company also showcased its latest achievements in multimodal interaction, aiming to accelerate innovation in intelligent agents, physical AI, and healthcare fields.

SenseNova U1.5-Lite-Preview by SenseTime: 8B Model Supports Native 4K Image Generation

Related Recommendations

SenseNova-Vision by SenseTime Open-Sources a Unified Vision Large Model, Achieving Outstanding Performance in Four Core Visual Tasks with a Single Model

16GB Memory, Local Instant Response! Google Releases Gemma 4 12B Revolutionary Encoder-Free Architecture Ignites Open Source Community

Google Launches New Gemma 4 12B Model: Easily Handle Visual and Audio Data Without an Encoder

New Developments in Domestic Large Models: MiniMax Launches the '10x Team' Program to Reward Global Top Experts

NVIDIA Releases the Nemotron 3 Series Open-Source Models: Inference Efficiency Surges 5 Times