Vidu Q4 Preview is the latest flagship AI video model from Shengshu Technology, a Tsinghua-born generative AI company, built for filmmakers, advertisers, and short-drama studios who need believable acting and cinematic control. Shengshu released it on October 7, 2026, and it targets teams that judge video on faces and performance rather than only on wide shots.
Core Features
- 4K output with 10-bit color and cinematic camera language, up from the 1080p ceiling of earlier Vidu models.
- Reference-driven generation: combine up to 15 reference images and 3 audio clips to lock characters, outfits, and voices across a series.
- Native audio: synchronized dialogue and sound effects generated with the video, with automatic camera switching.
- Lifelike performances: facial expression, emotion, body movement, and voice stay in sync, tuned for talking-head and UGC clips.
- Up to 16-second clips via image-to-video and reference-to-video endpoints, with no text-to-video in the preview.
Use Cases / Best For
- Advertising and e-commerce teams testing many creative versions fast.
- Short-drama and film crews previsualizing performances, blocking, and lensing.
- Creators who start from a real product shot or approved character image.
Pros / Cons
- Pro: 4K with native audio and up to 15 reference images for tight character control.
- Pro: Launch price around $0.014/sec makes high-volume testing cheap.
- Con: No text-to-video mode; every clip needs a source image.
- Con: Still a preview, so specs and pricing may change before the full release.
Pricing
Launch pricing starts at about $0.014 per second (RMB 0.09) at the lowest resolution and scales to roughly $0.39 per second at 4K, around $7.20 per minute at production settings. Image-to-video and reference-to-video generation carry a 30 percent discount through November 30, 2026. Vidu Q4 Preview debuted third on the Artificial Analysis image-to-video leaderboard at 1,179 Elo.
Our Take: Best for performance-driven short-form video where faces matter; the trade-off is the missing text-to-video mode, so every clip needs at least one source image.
See more AI Video tools.
