How to Use Video References for Precise Camera Movement

Why Video References Matter So Much

Text descriptions of camera movement are often imprecise. “Slow push in” means different speeds and distances to different people. Seedance 2.0’s video reference feature lets you upload a reference clip directly — the model understands and reproduces camera language accurately.

Choosing the Right Reference Video

Ideal reference videos should have:

• 3–10 seconds duration with clear, complete camera movement
• Single camera action — avoid complex multi-shot cuts
• Stable footage — minimize shake and rapid zoom
• Composition similar to the target scene for easier transfer

Multi-Clip Reference Strategy

Seedance 2.0 supports up to 3 video references. Recommended division:

1. Clip 1: Primary camera reference (push, follow, etc.)
2. Clip 2: Action rhythm reference (e.g., walking speed)
3. Clip 3 (optional): Transition or special effect reference

Case Study: Reproducing Cinematic Follow Shot

Suppose you want a follow shot of someone running through a forest:

Reference video 1: Upload a handheld follow shot of running
Reference video 2: Upload tone/style reference
Text: "An explorer sprints through a misty forest at dawn. Camera follows closely, occasionally passing through branches. Sunlight filters through the canopy."
Output: 16:9, 1080p, 12 seconds

Common Mistakes

• ❌ Reference too long or containing multiple unrelated shots
• ❌ Low-quality reference where the model can’t extract camera data
• ❌ Text contradicts reference (e.g., reference is push-in, text asks for pull-out)

Advanced: Camera + Image Combo

Combining video reference (camera) with image reference (visual style) is one of Seedance 2.0’s most powerful workflows. Use images to lock visual style, video for camera movement, and text for final detail.