China's Shengshu and Tsinghua drop Vidu to challenge Sora

Ever since OpenAI shocked the world with Sora, everyone has been waiting to see when a credible challenger would emerge from China.

At the 2024 Zhongguancun Forum in Beijing, Tsinghua University and Shengshu Technology threw their hat into the ring, unveiling Vidu—a text-to-video AI capable of spitting out 16-second clips at 1080p resolution with a single click.

16 seconds of 1080p, powered by U-ViT

While Sora grabbed headlines with its ability to generate up to 60-second clips, Vidu caps out at 16 seconds for now. However, hitting 1080p resolution out of the gate is nothing to sneeze at.

Under the hood, Vidu runs on a Universal Vision Transformer (U-ViT) architecture. The team claims they developed this architecture back in September 2022, which technically predates the Diffusion Transformer (DiT) framework that powers Sora.

According to Shengshu Technology, this setup allows Vidu to simulate the physical world with impressive accuracy, handling complex lighting, realistic shadows, and subtle facial expressions, while also hallucinating surreal and imaginative scenes.

Cinematic camera angles, but Sora still leads

One of the standout features shown off is multi-camera view generation. Vidu can fluidly transition between long shots, medium shots, and close-ups inside a single continuous scene, giving it a surprisingly cinematic feel.

The team boldly recreated several scenes originally used in Sora’s debut demo to prove its capabilities. The temporal consistency is clearly there—objects and people don't just melt into goo when the camera pans.

That said, in a direct side-by-side comparison, Vidu still falls short of Sora's sheer visual fidelity and lifelike realism. It looks great, but Sora's film-grade polish still holds the crown.

Why it matters

Vidu marks China's first high-profile, direct answer to OpenAI in the advanced text-to-video race. For creators, generating 16 seconds of multi-angle 1080p footage with good temporal consistency could soon be a massive win for rapid storyboarding and indie concept art, even if public access is still in rollout mode.

From an industry lens, the U-ViT architecture is the real talking point. It shows that Chinese frontier labs aren't just copying Western playbooks; they have native architectural foundations that trace back years.

We think Shengshu's decision to directly recreate Sora's exact demo prompts was both confident and risky. It clearly demonstrates how close the gap has gotten, but it also makes the remaining fidelity gap impossible to ignore. Still, if this is where their baseline sits, the video generation race just got a lot more interesting.