GitHub repositories for TurboDiffusion, SageAttention, SLA, and a technical report, along with model checkpoints for image-to-video and text-to-video generation, are provided.

In practical tests, the Vidu model can generate a 1080p, 8-second video in about 8 seconds after acceleration, dramatically cutting production time from hundreds of seconds.

SageAttention enables low-bit attention acceleration and has been deployed at scale across GPUs from various vendors, with adoption by Tencent Hunyuan, Alibaba Tora, Baidu PaddlePaddle, Google Veo3, SenseTime, and vLLM.

The TurboDiffusion framework, developed by ShengShu Technology and TSAIL Lab, delivers real-time AI video generation with claimed 100× to 200× speedups on a single RTX 5090, while maintaining minimal loss in visual quality.

TurboDiffusion is presented as a milestone-level technology with broad applicability, aimed at accelerating real-world adoption of video-generation AI for creators and enterprises.

Users can install the TurboDiffusion Python package, download model checkpoints such as TurboWan2.1-T2V-14B-720P, and run the provided scripts to generate content.

The release emphasizes ongoing innovation to boost efficiency, cut costs, and accelerate real-world adoption of generative AI in creative industries.

SageAttention, a world-first low-bit attention acceleration, has been adopted in NVIDIA TensorRT and other platforms, and is employed by major industry players and researchers.

ShengShu Technology and Vidu are positioned as leaders in multimodal AI and video generation, with a global footprint across multiple industries and partners.

Open-source resources and project links are provided for TurboDiffusion, SageAttention, SLA, and rCM, along with information about Vidu and ShengShu Technology.

Direct links to TurboDiffusion and related technologies, plus ShengShu’s Vidu products, are available and indicate deployment across multiple platforms and industries.

The release is framed as a turning point for the industry, described as a DeepSeek Moment for video foundation models, signaling a path to scalable, real-time commercial deployment.