All models

AlibabaVideo

Wan 3.0

Alibaba's Wan 3.0, smooth motion and native audio from a prompt, a frame or references.

video/wan-3
Try it in StudioUse with your agent or API

The tiny figure on the palm waves at the camera, does a happy little dance with a spin and finishes with a bow; the hand bobs gently and the camera slowly pushes in. One continuous shot, no cuts, smooth confident camera, consistent character, correct anatomy, no text, no logos, no captions.

Price
10 credits / second
Makes
Text to video, Image to video, Reference to video
Clip length
5 to 15 s

Best for

Wan 3.0 is the successor to Alibaba's Wan 2.7 video model. It ranked second for text to video on the Artificial Analysis video arena in September 2026. In Summer it renders 720p clips with native audio from a prompt, a start frame with an optional end frame, or up to ten reference images.

  • Cutscene shots with ambient sound in one pass.
  • Turning concept art into a moving establishing shot.
  • Consistent character shots from a few reference images.

Put it together

Where Wan 3.0 fits in a game, with the models it works well with. Your agent runs every step.

Price

ModeModelPrice
Text to videoWan 3.010 credits / second
Image to videoWan 3.010 credits / second
Reference to videoWan 3.0 References10 credits / second

Paid from your Summer credits.

Use it from code or your agent

Sign in once with summer login. Your agent or your code then runs it with your paid credits.

npx -y summer-engine@latest login
npx -y summer-engine@latest api run video/wan-3 \
  --input '{"prompt":"a dragon circling a castle at dawn","duration":5}' \
  --download .

video/wan-3: $0.10 per second of video. Paid from your Summer credits.

Inputs

What input takes for video/wan-3. Send only what you need; Summer fills in tested defaults.

  • promptstringup to 4000 characters

    What happens in the clip. Needed without imageUrl.

  • imageUrlstringup to 2048 characters

    A public image URL to animate (image to video).

  • endImageUrlstringup to 2048 characters

    With imageUrl: the last frame. The same image as imageUrl makes a seamless loop. Models with an end frame: Seedance, Kling 3, Wan, Vidu, LTX 2.5, Luma, MiniMax.

  • referenceImageUrlsarray of string1 to 10 items

    Reference to video: public images of the characters, props or style to keep (with a reference model, e.g. kling-4-ref). Motion transfer: the character to move.

  • videoUrlstringup to 2048 characters

    An input clip (public URL, at most 20 s): restyle or reframe it, move a character with its motion, or add sound to it (e.g. kling-video-to-audio). Billed on the clip's length.

  • durationintegerDefault 5 · 1 to 20

    Seconds; the model renders its nearest supported length.

  • aspectRatioenumDefault 16:9
    • 16:9
    • 9:16
    • 1:1
    • 4:3
    • 3:4
    • 21:9
    • 3:2
    • 2:3
    • auto
  • negativePromptstringup to 1000 characters

    What the clip must not show.

  • resolutionenum

    Only a quality the model offers; leave out for its default.

    • 720p
    • 1080p
    • 4k
    • 480p
    • 768p
    • 2k
    • 768P
    • 1080P
    • 480P

Model docs · Summer API docs · Set up the Summer MCP server · Try it in the browser

Questions

What is Wan 3.0 good for in game development?
Wan 3.0 is the successor to Alibaba's Wan 2.7 video model. It ranked second for text to video on the Artificial Analysis video arena in September 2026. In Summer it renders 720p clips with native audio from a prompt, a start frame with an optional end frame, or up to ten reference images. Typical game uses: cutscene shots with ambient sound in one pass; turning concept art into a moving establishing shot; consistent character shots from a few reference images.
How much does Wan 3.0 cost in Summer?
In Summer Studio: Wan 3.0 (text to video): 10 credits / second; Wan 3.0 (image to video): 10 credits / second; Wan 3.0 References (reference to video): 10 credits / second. It is charged from your Summer credits.
What can Wan 3.0 generate, and how long are the clips?
In Summer, Wan 3.0 supports text to video, image to video, and reference to video. Generated clips run 5, 8, 10, and 15 seconds, in 16:9, 9:16, 1:1, 4:3, and 3:4 aspect ratios at 720p and 1080p.
Can I use Wan 3.0 from Claude Code, Cursor or Codex?
Yes. Connect the Summer MCP server to Claude Code, Cursor, Codex or another MCP client, then ask your agent to generate with summer_generate_video with model: "wan-3". Setup instructions are at summerengine.com/mcp.
Where do Wan 3.0 results go?
Every generation is saved to your Summer asset library. From there you can download it or bring it into a Summer Engine project.