Shengshu Technology and Tsinghua University in China have introduced Vidu, a text-to-video model that can produce 16-second, 1080p resolution videos with just one click. This announcement was made at the 2024 Zhongguancun Forum in Beijing, where Vidu was presented as a formidable rival to OpenAI’s Sora. Vidu, like Sora, can generate 16-second clips at 1080p resolution.

Vidu is built on a Universal Vision Transformer (U-ViT) architecture, which, according to the company, enables it to simulate the real world with multi-camera view generation. This architecture was reportedly developed by the Shengshu Technology team in September 2022, predating the diffusion transformer (DiT) architecture used by Sora.

The company claims that Vidu can generate videos with complex scenes that adhere to real-world physics, including realistic lighting and shadows, and detailed facial expressions. The model also exhibits a rich imagination, creating surreal content that does not exist, with depth and complexity. Vidu’s multi-camera capabilities allow it to generate dynamic shots, seamlessly transitioning between long shots, close-ups, and medium shots within a single scene. During its demo, the company attempted to recreate scenes similar to those shared by OpenAI during Sora’s release. While Vidu is an impressive achievement and a testament to China’s rapid progress in AI research, a side-by-side comparison with Sora shows that the videos generated by Vidu do not match Sora’s level of realism. The output, while impressive, lacks in terms of visual fidelity.

However, it’s important to note that the temporal consistency achieved by Vidu is commendable, and this technology has the potential for further refinement and improvement over time. We will continue to follow on Vidu and its progress in the future as it becomes available.

Thank you for watching!

Keywords: #ai #texttovideo #artificialintelligence