Vidu AI

Tutorials

How to Configure Vidu Q3 Reference-to-Video: From Single Subject to Multi-Subject Mix

12 min Advanced Tutorial

Reference-to-Video (R2V) is a key Vidu Q3 capability for controllability: the model can learn character appearance, motion style, camera movement, and overall visual style from reference images or videos, then maintain higher consistency in later shots.

1. Decide consistency priorities first

Before uploading assets, align team priorities internally:

  • Character consistency first: for web comics, virtual idols, and brand IP.
  • Scene continuity first: for product demos and store walkthroughs.
  • Stylization first: for art films and visual experiments.

2. Single-subject reference: minimum viable loop

  1. Choose a reference image or short clip with a clear subject and minimal occlusion.
  2. In the prompt, explicitly state elements to preserve from the reference (hair, outfit, posture, etc.).
  3. Run an ~8 second test shot first to confirm composition and light direction.
  4. Then extend to a full 16 second shot—avoid starting with long takes that require rework.

Tip: Keep reference video pacing stable; avoid heavy camera shake, or the model may overfit to jitter when learning camera motion.

3. Multi-subject mix: up to 7 reference subjects

When you need multiple characters in frame or product + spokesperson together:

  • Give each subject its own close-up reference when possible;
  • Label explicitly in the prompt, e.g. “left character = reference A, right character = reference B”;
  • For strong dialogue and lip-sync needs, specify language and tone clearly.

4. Combine with start-end frame control

If you need a harder constraint on opening and closing frames, see How to Use Vidu Q3 Start-End Frame Control for Transition Shots.

5. Further reading: understand model limits from benchmarks

To understand Vidu Q3’s strengths from benchmarks and rankings, read Artificial Analysis Review Explained: Why Vidu Q3 Outperforms Runway and Veo.

Notes

  • Low-resolution references may “average out” fine texture—prefer clear source assets.
  • For serialized production, maintain a reference asset version table so the team doesn’t mix old references.

After you build 2–3 reference templates, package them as internal “style packs” in your shared workbench to cut multi-platform switching cost.