Reference to video
Reference to Video AI
Give the model pictures, clips and audio to follow — a character, a style, a camera move, a voice — and say in the prompt how each one is used.
- Up to 20 files
- 10 images · 5 clips · 5 audio
- Type @ to name them
- Sound included
- Up to 30 s
Reference-to-video examples
Real renders from this site, with the exact prompt. Use one as a starting point.
“A red paper boat drifts down a rain-soaked city gutter at dusk, neon reflections ripple on the water, the camera follows low and close, soft rain and distant traffic”
Text · Wan 3.0 · 720P · 5s
Use this prompt“Handheld walk through a crowded night market in Taipei, steam rising from food stalls, a vendor flips skewers and smiles at the camera, sizzling grills and chatter”
Text · Wan 3.0 · 1080P · 5s
Use this prompt“A teenage skater carves down an empty sunset street, the camera tracks beside her at board height, wheels rumbling on asphalt, warm lens flare”
Text · Wan 3.0 · 720P · 5s
Use this prompt“Top-down macro of a barista pouring latte art, milk folding into a rosetta, slow motion, quiet café ambience and a spoon clinking”
Text · Wan 3.0 Prime · 720P · 5s
Use this prompt“Drone shot gliding low over a turquoise glacier lake at sunrise, mist drifting across the water, a slow rise reveals snow peaks, deep wind and ambient strings”
Text · Wan 3.0 · 480P · 10s
Use this prompt
Guide the video with references
References are files the model follows without using them as frames: a character's face and outfit, the look of a scene, the rhythm of a camera move, the voice or music you want.
Each upload gets a name in order (Image 1, Image 2, Video 1, Audio 1), and the site passes them to Wan 3.0 under the labels it expects. Type @ in the prompt to insert one, then say what it's for: "Image 1 is the main character, follow the camera move in Video 1".
Clips must be 2–15 seconds each and 15 seconds in total per kind, and the output plus reference video can't be longer than 30 seconds. Reference video is billed per second like output; reference images are free. Reference mode works on Wan 3.0 and Wan 3.0 Prime.
Wan 3 API is an independent service and is not affiliated with or endorsed by Alibaba. Wan and Wan 3.0 are trademarks of their owners.
How it works
Three steps, a couple of minutes, no GPU.
- 01
Describe the shot
Say what happens: subject, action, camera, light and sound. Or start from an example.
- 02
Add pictures or references
Optional: a first or last frame, or reference images, clips and audio to follow.
- 03
Generate and download
Pick the model, resolution and length, see the price, and download the MP4 when it's ready.
More ways to create
Wan 3 API
One REST API for Wan 3.0, Prime and Wan 2.7, with a playground.
Wan 3.0
Alibaba's newest model: up to 30 s with sound.
Wan 3.0 Prime
Alibaba's premium Wan 3.0 tier, for final renders.
Wan 2.7
Previous generation: 720P or 1080P, 2–15 s.
Text to video
Describe a shot and get a video with sound.
Image to video
Animate a picture, or move from one image to another.
Reference to video questions
Choose Reference above, add images, clips or audio, then describe the new video and what each reference is for. The model keeps what you point it to and generates the rest.
By upload order: Image 1, Image 2, Video 1, Audio 1. Type @ in the prompt box to pick one, or use the chips under the box.
Up to 10 images, 5 video clips and 5 audio clips. Clips are 2–15 seconds each and 15 seconds in total per kind; audio needs at least one image or clip.
Reference video seconds are billed at the same rate as output seconds. Reference images are free. The button shows the total before you generate.
Image to video uses your picture as the actual first (and last) frame. Reference mode uses pictures and clips as guides, so the video can show the character or style in a brand-new shot.
They're used only to make your video and deleted 30 days later. Use files you own or have permission to use.
Make your first Wan 3.0 video
Write a sentence, pick a length, and hear it come to life.
Start creating