Video extend takes a clip you already have and continues it. You supply the starting video and describe what should happen next; the model carries the shot forward rather than starting over. It is the tool for a moment that needs a few more seconds, not for building a whole sequence from scratch.
The model sees the last frames of your clip, so the prompt should read as the next beat, not as a fresh brief. If the clip ends with someone reaching for a door handle, "the door opens and they step through into daylight" continues it. Re-describing the earlier part of the shot works against you — the model already has that information.
Keep the camera language consistent with what the clip is already doing. A clip that ends mid-push-in should keep pushing, not cut to a new angle; if you want a different angle, that is a new shot rather than an extension.
It suits a shot that is almost long enough: a product reveal that lands a second too early, a mood clip that needs to breathe. It is a poor fit for changing what the clip is about, for adding a cut to a different location, or for repairing a clip whose motion is already wrong — those are better solved by regenerating the shot with text to video or image to video.
Treat the first extension as a test. Quality varies by route and by how complex the continuation is.
How much longer can I make a clip? Duration limits and the resulting credit cost are shown in the form before you submit.
How much does it cost? The credit cost appears beside the form and depends on the model, resolution and duration. Credit prices are on pricing.
Where do I start? Open the Wan workspace and choose the video-extend tab.