Reference to video lets you supply example footage alongside your prompt, so the result follows the look of that footage rather than an invented style. You describe what should happen in the new clip; the reference clips carry the visual treatment — grading, texture, motion feel.
The reference clips should resemble one another. Two clips from different sources, with different grades and different motion, pull the output in two directions and the result usually settles somewhere unhelpful.
Because the look is coming from the references, the prompt should carry the content: who or what is in the new shot, what it is doing, and how the camera behaves. "A delivery van turns into a wet car park, handheld, following the vehicle" is a content instruction; leave the grade to the references.
If an artefact keeps appearing, name it in the negative prompt rather than adding more description.
Use it when a series needs to look consistent — several shots for one project, or a new scene that has to sit alongside footage you already have. It is not a way to copy a specific person, brand or copyrighted work, and it cannot invent a style it has never been shown. Treat the first render as a test: how closely the output tracks the references varies by route and by how similar your references are.
How many reference clips should I add? One or two, and they should look like each other.
How much does a clip cost? The credit cost appears beside the form before you submit and depends on the model, resolution and duration. Credit prices are on pricing.
Where do I start? Open the Wan workspace and choose the reference-to-video tab. If you have no footage to reference yet, start with text to video instead.