🌟 Want to try SenseNova U1 yourself? The model is open-source, free to use, and ready to run today. Give it a complex prompt. Watch it generate text AND images in one shot. I promise it'll surprise…

They also open-sourced an 8-step distilled LoRA for ComfyUI deployment: 100 NFE → 8 NFE H100 inference time: 23s → 2s ComfyUI is now supported too, with ready-to-run workflows for text-to-image,…

商汤原生多模态统一模型 SenseNova-U1 这就是 SenseNova-U1，商汤 SenseNova 团队首创 NEO-unify 架构的原生多模态统一模型！ 彻底干翻传统图文适配器+VAE老路： 单一架构原生统一理解、推理、生成，语言和像素直接端到端建模，抛弃视觉编码器 + VAE，从第一性原理重构多模态范式，信息密度、风格一致性、图文连贯性直接拉到新高度！

最近对摄影有点感兴趣😂 我就随手丢给它一句：“我想学习一下风景摄影构图，但是不会用相机，可以给我讲讲索尼相机的使用吗？” 结果它直接给我整了一套“小白友好版摄影教程”。

Commercial-grade power, compact scale. We’re open-sourcing the U1 Lite series: - U1-8B-MoT (Dense) - U1-A3B-MoT (MoE) Leading the SOTA for open-source efficiency in reasoning, spatial intelligence,…

商汤最近开源了一款专用于信息图生成的 AI 模型，效果特别惊艳。 可一键生成海报、图表、食谱卡片、明信片，甚至是 arXiv 风格的页面。 官方放出了 100 多个生成案例及提示词，涵盖各种风格的信息图，可以直接参考学习。 GitHub： 模型下载： 技术报告： 项目基于 SenseNova-U1 原生多模态模型微调优化，在文字渲染和图表准确性上明显提升了很多。…

Open-sourced an 8-step distilled LoRA: 100 NFE → 8 NFE, cutting H100 inference from 23s to 2s. ComfyUI is now supported, with ready-to-run workflows for t2i, image editing, and interleaved generation

The frontier NMM research coming out of China right now is genuinely world-class. SenseNova U1's technical report is the most complete open NMM methodology published by any lab:

Why this is a big deal: • SOTA efficiency in compact models (8B & MoE) • Native image–text interleaving • Strong reasoning + spatial understanding • Built for real-world use cases

Results: