Wan 2.1
Make video clips from a text prompt or a start image with Wan 2.1 by Alibaba, from 368 credits per clip. See its settings, limits and exact prices.
For the complete documentation index, see llms.txt. This page is also available as Markdown.
Wan 2.1 is an Alibaba video model, tagged Fast and Open Source. It makes a 5-second clip from a text prompt or a start image, at 16:9 or 9:16, in 480p or 720p, with no sound option. The price follows the resolution: 368 credits at 480p and 736 at 720p.
#At a glance
- Maker
- Alibaba
- Type
- Video model
- Modes
- Text to video and image to video
- Aspect ratios
- 16:9, 9:16
- Outputs per run
- 1 to 4, set with Count
- Start image
- Optional start image (JPEG, PNG or WEBP, up to 20 MB), one image only.
- Length
- 5 s only
- Resolution
- 480p, 720p
- Sound
- No sound option
#What it costs
These are credits per clip, as of 2026-10-10. The Studio shows the live quote for your settings before you generate, and the total it charges is that quote multiplied by Count.
| Length | 480p | 720p |
|---|---|---|
| 5 s | 368 | 736 |
#Use it in Studio
Step 1: Open Create Video
In Studio, open Create Video. Sign in when the Studio asks, because every generation needs an account.
Step 2: Choose Wan 2.1
Open the model picker, which groups models by maker, and choose Wan 2.1.
Step 3: Write your prompt
Describe the clip. To start from a picture, attach one start image (JPEG, PNG or WEBP, up to 20 MB).
Step 4: Set the output
Set aspect ratio, resolution and Count from 1 to 4.
Step 5: Check the price and generate
The Studio shows the credit price for your settings. Videos run in the background: the card updates in place when the clip is ready, and the clip is saved to your Library.
#Limits
- You need to be signed in to generate. Wan 2.1 is available on every plan, including Free; credits are the limit.
- One start image only. The Studio has no end frame, no multiple reference images and no video-to-video input.
- A video that has not finished within 30 minutes is marked failed and refunded. A failed generation is refunded, and the Studio does not run a different model in its place.