Wan 3.0 Pro reference-to-video holds a face, garment, set, movement or voice using image, video and audio references, plus an optional document or web link. It renders 2 to 30 seconds at 1080p, 2K or 4K.
Workbench
These examples were written for this endpoint. Replace the prompt and sample media URLs with your own accessible inputs, then check the current quote and model availability before submitting.
- Output8s
same-two-three-times Two reference photos, one couple, three times of day in eight seconds: coffee by the window in the morning, cooking together at dusk, one blanket on the balcony at night. The test is whether both faces stay the same two people across three cuts and three lighting setups.
- 1080p
- 16:9
- 8s
- Output8s
face-and-set The prompt spends one clause on identity — 'keep her face, hair and the black gown' — and the rest on one action: walk from the doorway to the window and stop. `reference_image_urls` gives the woman first, the room second; `duration_seconds` `8` is entirely that walk and the slow lateral track after it. `seed` `640117` at `resolution` `1080p` keeps the room reproducible.
- 640117
- 1080p
- 16:9
- Output6s
garment-hero-4k `reference_image_urls` here is a flat packshot of the coat alone, no model — the prompt invents the wearer while holding 'cut, colour and stitching' exact, harder than matching a photo of someone already in it. At `resolution` `4k`, buttons and topstitching stay crisp, and the hem visibly lifts and swings open on the turn. `generate_audio` is off, `duration_seconds` is `6`.
- 4k
- 6s
This example starts at $0.144 per output second. The Playground updates the estimate when you change the settings.