Reference to Video

Build a small, non-conflicting reference set

Give each file a reason to exist—identity, wardrobe, prop, environment, motion or sound—and remove references that disagree on a critical detail. WAN3 accepts up to twenty mixed references, including up to five videos. Wan 2.7 accepts up to five image or video references.

• Identity reference: face, silhouette or product shape.
• Appearance reference: wardrobe, material, palette or surface detail.
• Environment reference: location, lighting and spatial context.
• Motion reference: pace, gesture or camera behavior.
• Audio reference: voice direction only when the shot needs it.

Resolve conflicts before uploading

Compare the references for identity, color, scale, lighting and camera perspective. If two files give opposite answers, choose the one that matters most instead of hoping the model will merge them. A reference with no assigned role should be removed from the first test.

Check supported files before uploading

Use WAN3 when audio belongs inside the mixed reference set or the output needs up to 30 seconds. Use Wan 2.7 for a 2–10 second result guided by image or video references. In WAN3, a reference video's rounded-up duration plus the selected output duration cannot exceed 30 seconds.

Compare duration, quality and credit rates

Credits depend on the selected model, resolution, duration and references. The studio shows the total before generation.

Write the prompt around relationships and action

The references already show appearance, so use the prompt to explain which subject is present, what happens, how the camera observes it and what must remain consistent. Keep the first attempt to one scene and one main action, then change one reference or instruction at a time.