How to Use Vidu Reference-to-Video for Characters
Vidu is an AI video generator from Shengshu Technology and Tsinghua University. Its Reference-to-Video mode accepts up to seven reference images and reuses them as recurring characters or props, which solves the consistency problem that breaks most AI video workflows.
Consistency is the reason studios give up on AI video. A character that changes face between shots makes a multi-scene story impossible. This feature targets that directly.
How Do You Use Reference-to-Video in Vidu?
- Open the Reference-to-Video mode in the workspace
- Upload up to seven images — one per character or prop
- Name each reference so prompts can call them by role
- Write a scene prompt referring to those names
- Generate and check that each subject stays recognizable
- Reuse the same reference set across every shot in the sequence
The reference set is your cast list. Keep it unchanged across scenes and the characters hold; swap an image mid-project and the continuity breaks.
What Makes a Good Reference Image?
Front-facing, sharp, evenly lit, and with the subject filling a good portion of the frame. A cropped, blurry, or side-lit reference gives the model too little to anchor on. If you only have one usable photo of a character, expect drift on profile angles.
Is Vidu Free?
There is a free tier with limited credits, and paid plans raise generation volume and resolution. Reference-to-Video and the fast lower-latency mode both consume credits, so a multi-shot sequence adds up quickly.
Other Generation Modes Worth Knowing
- Text-to-video: prompt to clip in several aspect ratios
- Image-to-video: animate a still with pan, zoom, or dolly
- Fast mode: lower latency for drafting and iteration
- Character consistency: the headline mode for episodic content
How Does It Compare to Kling?
| Capability | Vidu | Kling AI |
|---|---|---|
| Reference characters | Up to 7 images | No equivalent |
| Max clip length | Standard tier | Up to 2 minutes |
| Motion realism | Good | Stronger on human motion |
| Best for | Episodic, recurring characters | Long single shots |
If your project needs the same person across ten shots, Vidu is the more direct tool. If you need one long, physically convincing shot, a longer-clip model wins.
What Still Goes Wrong?
Extreme angles break consistency, and scenes with three or more referenced subjects get crowded. Sometimes the model applies the right face to the wrong body. Keep scenes to two characters where possible and verify each clip before building the next one.
Quick Summary
- Vidu Reference-to-Video keeps up to seven subjects consistent across clips.
- Keep the same reference set across the whole sequence.
- Use front-facing, well-lit, uncropped reference images.
- Fast mode is for drafting before you commit credits.
- Extreme angles and crowded scenes still break consistency.
