Consistent characters
Keep the same face and outfit across the clips of a series or a short drama.
Wan3 AIWan 3.0 Reference to Video Generator
Overview
What Is Wan 3.0 Reference to Video?
Reference to video gives Wan 3.0 pictures to follow instead of frames to copy. You upload images of a face, an outfit, a product or a place, and the model keeps their look while it builds a new scene from your prompt.
You point at each image in the prompt. On Wan3 AI you type @image1, @image2 … in upload order; we pass them to Wan 3.0 as Image 1, Image 2 …, the naming Alibaba's guide uses.
Alibaba's API also accepts reference videos, audio, documents and web pages. Wan3 AI currently offers reference images only, up to 9 per clip. Wan3 AI is an independent site, not an Alibaba product. We call Wan 3.0 through a third-party API and show the credit cost of every clip before you generate it.
Reference to video on Wan3 AI
- Input
- 1–9 reference images plus a prompt
- Naming
- @image1 … @image9, in upload order
- Image files
- JPG, PNG or WEBP, up to 10 MB each
- Clip length
- 2–30 seconds
- Resolution
- 480p, 720p or 1080p
- Audio
- Generated with the video; can be switched off at the same price
- Not offered here
- Reference videos, audio, documents and web pages
- Lowest cost
- 14 credits (5 s at 480p)
Facts checked on 11 October 2026 against Alibaba's documentation.
How it works
Reference to Video in Three Steps
- 01
Upload your references
Add up to 9 images: the character, the outfit, the product, the setting. One subject per image keeps things clear.
- 02
Name them in the prompt
Type @ to pick an image and say what each one is for, for example "@image1 wears the jacket from @image2 in the café from @image3".
- 03
Set the format and generate
Choose length, resolution and aspect ratio, check the credit cost and generate.
Use cases
What to Make With Reference to Video
Product ads
Put your real product, from your own photos, into new scenes and settings.
Style and location
Use a style frame or a photo of a place to set the look of the scene.
Multi-subject scenes
Bring several people, objects and a background together in one shot, each from its own image.
Prompt tips
Get More From Your References
One job per image
Give each upload one clear role (the face, the jacket, the room) and use clean, well-lit pictures.
Name every image
Mention each upload by its @image number and say what it is for; unnamed references are easy to mix up.
Still describe the scene
The references set the look; the prompt still has to say where we are, what happens and how the camera moves.
FAQ
Wan 3.0 Reference to Video FAQ
How many reference images can I use?
Up to 9 per clip on Wan3 AI. Alibaba's API allows up to 10.
How do I refer to an image in my prompt?
Type @ in the prompt box and pick an image, or write @image1, @image2 … in upload order. We send them to Wan 3.0 as Image 1, Image 2 …
Can I use reference videos or audio?
Not yet. Wan 3.0 can read reference videos, audio, documents and web pages, but Wan3 AI currently supports reference images only.
Can I combine references with a first or last frame?
No. Wan 3.0 does not accept reference images and first or last frames in the same request. Use Image to Video when the clip has to start or end on a specific picture.
How much does it cost?
A 5-second clip costs 14 credits at 480p, 28 at 720p and 55 at 1080p; 30 seconds at 1080p costs 329. The form shows the exact cost for your settings before you generate, and failed generations are refunded automatically.
Build a Scene From Your References
Upload your images, name them in the prompt and check the credit cost before you generate.