Live now, from $0.05 per second

Wan3.0 and 3.0 Prime are live on Runware

Alibaba's all-in-one multimodal video model. Text, images, video, audio, documents, and webpages go in, and up to 30 seconds of 1080p video with native audio comes out. Wan3.0 Prime runs the same model with lower latency.

Try in PlaygroundSee pricing
0:00

Every workflow in one model.

Generation, reference-driven video, and editing in one multimodal model, through one Runware integration.

  • Text-to-video
  • Image-to-video
  • First and last frame
  • Reference images
  • Reference video
  • Audio-driven video
  • Documents and webpages
  • Video editing
  • Extension

The Wan3.0 lineup

Wan3.0 is the full model. Wan3.0 Prime runs the same model at lower latency.

0:00
Live on Runware
alibaba:[email protected]

The full multimodal model.

From $0.05 per second at 480p

  • 2 to 30 seconds in a single pass, or auto
  • 480p, 720p, and 1080p in five aspect ratios
  • Up to 10 reference images, 5 videos, and 5 audio clips
  • First and last frames, documents, and webpages as input
  • Native audio, editing, and extension
Try in PlaygroundDocs

Wan3.0 and 3.0 Prime pricing

Metered per second of output, by model and resolution, whatever the inputs. No subscription, no minimums, and the same key and billing as every other model on Runware.

480p
$0.05/sec
720p
$0.10/sec
1080p
$0.20/sec

Every input type and workflow, at the same rate per second whatever goes in.

Ranked #1 for text-to-video

Artificial Analysis ranks video models by blind human votes between pairs of clips, scored as Elo. On its text-to-video arena, which judges video with audio, Wan3.0 is first on the whole board, and 139 points above Wan2.7.

  1. Wan3.01175
  2. MiniMax H31161
  3. Seedance 2.51155
  4. Gemini Omni Flash 1.21154
  5. FLUX 3 Video1137
  6. Wan2.71036
ArenaWan3.0MiniMax H3Seedance 2.5Gemini Omni Flash 1.2FLUX 3 VideoWan2.7
Elo score117511611155115411371036
95% confidence interval±13±13±13±14±13±14
Rank on the full board#1#3#4#5#10#19
Votes7,6517,8127,7655,2256,8276,303

Artificial Analysis text-to-video arena (AA-Video-T2V v2.1), as of November 12, 2026. Lines show each score's 95% confidence interval, so where two overlap, the gap is within the margin. Ranks are based on the full board, which includes models not on Runware.

See the full leaderboard

Native 30-second outputs

Wan3.0 generates up to 30 seconds in a single request, so a scene plays out with one continuous cast, camera, and soundtrack instead of stitched clips.

0:00
30 seconds

A 30-second oner on a follow cam, holding one hiker in a single unbroken take as an alpine meadow gives way to a mushroom forest, a path across the clouds, and the back of a flying whale.

0:00
30 seconds

A 30-second POV descent in one generation, from a snowy ridge at sunrise through scree and forest, down to alpine meadows and a turquoise lake.

What goes in, what comes out

Two of Wan3.0's input workflows, with the files that went into each request and the video that came back.

Inputs
  • First frame input
    First frame
  • Last frame input
    Last frame
Wan3.0 result
0:00

Pin the opening and closing frames, and Wan3.0 writes the motion between them: here, a scoop of ice cream that doesn't survive the trip.

Made on Runware

0:00
Text to video

Seaplane landing at sunrise

10 seconds · 1080p

Prompt

One continuous 10-second cinematic shot. A vintage red and cream single-engine seaplane descends toward a glassy alpine lake at sunrise, surrounded by pine forest and granite peaks with fresh snow on the summits. The camera starts low over the water, tracking alongside the plane from the left at the same speed as it glides down. The floats touch the surface and carve two long white wakes, throwing fine spray that catches the low gold sunlight. The perfect mirror reflection of the mountains breaks into ripples behind it. The plane decelerates and settles, propeller slowing, as the camera rises and arcs gently around the nose to a three-quarter front view with the sun flaring softly behind the peaks. Thin morning mist drifts across the lake. Physically accurate water, spray, reflections and aircraft motion. Premium travel-commercial cinematography, anamorphic lens, rich natural color, crisp detail. Sound: steady radial engine drone that eases off, the hiss and slap of floats meeting water, spray settling, a distant loon call, then quiet lake ambience. No people visible, no dialogue, no text, no logos.

How it compares

Wan3.0 next to Wan2.7 and the other leading video models on Runware. Wan3.0's edge is breadth: 30-second scenes, every reference type, and documents and webpages as input, with editing and extension in the same model.

ModelWan3.0Wan2.7MiniMax H3Seedance 2.5Gemini Omni Flash 1.2FLUX 3 Video
Max duration30s15s15s30s15s20s
Max resolution1080p1080p1440p1080p4K1080p
Reference inputs10 images, 5 videos, 5 audio clips1 image, 1 video9 images, 3 videos, 3 audio clips30 images, 10 videos, 10 audio clips10 images, 3 videosUp to 10 keyframe images
First and last frameYesYesYesYesYesYes
Documents and webpagesYesNoNoNoNoNo
Native audioYesYesYesYesYesYes
Edit and extendEdit and extendNoEditEditEdit and extendNo

Wan3.0 guides

Step-by-step guides to prompting Wan3.0, and to working with image, video, audio, document, and webpage inputs on Runware.

Prompting Wan3.0

Layered directives, camera language, style range, directing the native audio, and the dimension and duration rules.

Image inputs with Wan3.0

Pin a first frame, morph between two frames, and keep a subject or product consistent with reference images.

Video and audio references with Wan3.0

Carry a subject into new scenes with reference videos, drive a video from audio, and combine reference types in one call.

Document and web inputs with Wan3.0

Turn structured content into video, such as a recipe into a cooking video or a public webpage into a short montage.

Get started

One API call, in whichever language you're already using, with the same key and billing as every other model on Runware. For Wan3.0 Prime, change the model ID to alibaba:[email protected].

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "videoInference",
      "taskUUID": "<uuid-v4>",
      "model": "alibaba:[email protected]",
      "positivePrompt": "A slow tracking shot along the deck of a night ferry as a deckhand coils rope and looks out at the city lights, then turns to the camera and nods. Sound: engine hum, waves against the hull, a distant horn.",
      "width": 1280,
      "height": 720,
      "duration": 10,
      "includeCost": true
    }
  ]'

Response

[
  {
    "taskType": "videoInference",
    "taskUUID": "<uuid-v4>",
    "status": "success",
    "videoUUID": "8d2f4b6a-1c3e-4a5b-9f7d-2e6c8a0b4d13",
    "videoURL": "https://vm.runware.ai/video/ws/2/vi/8d2f4b6a-1c3e-4a5b-9f7d-2e6c8a0b4d13.mp4",
    "cost": 1
  }
]
Full API referenceMigrating from another provider?
Launch in Playground

Frequently asked questions

What is Wan3.0?

Alibaba's all-in-one multimodal video model. It takes text, images, video, audio, documents, and webpages, and returns up to 30 seconds of 1080p video with native audio. It also edits and extends existing footage.

How much does Wan3.0 cost on Runware?

Wan3.0 is billed per second of output: $0.05 at 480p, $0.10 at 720p, and $0.20 at 1080p, so a 10-second 1080p clip costs $2.00. Wan3.0 Prime is $0.068, $0.14, and $0.28.

What's the difference between Wan3.0 and 3.0 Prime?

Prime runs the same workflows at the same output quality, with lower latency, for a higher rate per second. Switch between them by changing the model ID.

How long can a Wan3.0 video be?

Any whole number of seconds from 2 to 30 in a single pass, or auto to let the model pick. To go further, extend an existing clip.

What inputs does Wan3.0 accept?

A text prompt of up to 20,000 characters, plus up to 10 reference images, 5 reference videos, 5 reference audio clips, a first and last frame, one document, and one public webpage.

How does Wan3.0 rank against other video models?

On Artificial Analysis's text-to-video arena, which scores video with audio by blind human votes, Wan3.0 ranks first with an Elo score of 1175, as of November 12, 2026. Its predecessor, Wan2.7, scores 1036.

Does Wan3.0 generate sound?

Yes. Audio is rendered with the video in the same request and is on by default. Set settings.audio to false for a silent clip.

Talk to us about Wan3.0

Planning volume on Wan3.0 or 3.0 Prime? Leave your details and our team will follow up, including on rate limits and dedicated capacity.