Vidu

规格参数

定价

Vidu is the flagship video generation model from Shengshu AI (生数科技), a Beijing-based AI research company with deep roots in character animation and generative media. What sets Vidu apart from the crowded field of image-to-video models is its unique hybrid architecture that combines video diffusion generation with character animation techniques — enabling persistent visual identity across all frames of a clip. This makes Vidu particularly powerful for animated characters, brand mascots, and any use case where a specific character must remain visually locked throughout the video.

能力说明

Vidu's core technical contribution is identity-preserving video synthesis: the model maintains a character's facial features, clothing design, colour palette, and body proportions with high fidelity even through dynamic actions and perspective changes. This capability is rare in AI video models, which typically show progressive drift in character appearance over longer clips. Vidu supports three aspect ratios and generates at up to 1080p, with clip durations up to 8 seconds — sufficient for social media content, game cinematics, and animated advertising spots.

最佳实践

Supply a clean, front-facing character reference image with minimal background clutter and consistent lighting to help Vidu build a stable identity representation. For stylised characters, use the original artwork or a rendered turntable frame as your input rather than a photo of a costume or real person. In your motion prompt, describe actions using animation vocabulary: "character waves cheerfully and jumps, landing with knees bent, eyes wide and mouth open in a smile" gives Vidu clear animation beats to follow. Avoid prompts that require the character to move off-screen and re-enter, as re-entry sequences are more prone to identity drift.