商汤科技发布了 Artist v1.0 Alpha，这是一款拥有 100 亿参数的文本到图像扩散模型，旨在与 Midjourney v…

SenseTime has launched Artist v1.0 Alpha, a 10-billion parameter text-to-image diffusion model designed to rival Midjourney v6 and Stable Diffusion 3 in prompt adherence and high-resolution generation. The model utilizes a Diffusion Transformer backbone scaled to 10 billion parameters, moving away from traditional U-Net architectures. By integrating a large language model text encoder, Artist v1.0 Alpha significantly improves compositional understanding. This allows the model to accurately render complex spatial relationships, multiple interacting subjects, and precise typography. SenseTime reports the model generates native 1024x1024 resolution images with robust photorealism and dynamic lighting. Internal benchmarks indicate it achieves state-of-the-art performance on the DrawBench and PartiPrompts evaluation suites, showing particular strength in cultural localization for Chinese prompts alongside standard English capabilities. The Alpha version is currently accessible via API through the SenseNova developer platform. This release signals an aggressive push by top-tier Chinese AI firms to capture the generative visual market and establish parity with leading Western multimodal systems.

商汤科技发布了 Artist v1.0 Alpha，这是一款拥有 100 亿参数的文本到图像扩散模型，旨在与 Midjourney v6 和 Stable Diffusion 3 在提示词遵循度和高分辨率生成方面相抗衡。 该模型采用了扩展至 100 亿参数的 Diffusion Transformer 主干架构，摆脱了传统的 U-Net 架构。通过集成大语言模型文本编码器，Artist v1.0 Alpha 显著提升了组合理解能力。这使得模型能够准确渲染复杂的空间关系、多个交互主体以及精确的排版文字。 商汤科技报告称，该模型可生成原生 1024x1024 分辨率的图像，具有稳健的照片级真实感和动态光照效果。内部基准测试表明，该模型在 DrawBench 和 PartiPrompts 评测套件上达到了最先进的性能，在针对中文提示的文化本地化方面表现出色，同时具备标准的英文能力。 Alpha 版本目前可通过 SenseNova 开发者平台经由 API 访问。此次发布标志着顶级中国 AI 公司正在强势进军生成式视觉市场，并力争与西方领先的多模态系统实现平起平坐。