Skip to main content

Wan 3.0

One endpoint for every mode: generate from a prompt, from a first (and last) frame, or from a mix of reference images, videos and audios that you cite in the prompt.
Wan 3.0 is Alibaba’s unified video generation model. A single request covers text-to-video, first/last frame image-to-video and reference-guided generation; the mode follows from the fields you send. It outputs MP4 videos from 2 to 30 seconds at 480p, 720p or 1080p, with a synchronized audio track (dialogue, sound effects, music) generated by default. It comes in two tiers with the same parameters: Wan 3.0, and Wan 3.0 Prime, the same model on Alibaba’s priority queue. Prime finishes a generation in about 2 minutes instead of about 14, at a higher price per second. Each tier has its own create, list and status endpoints.

​
Key capabilities

  • Three modes in one endpoint: Text-to-video, first/last frame, and reference-guided generation
  • Mixed references: Up to 10 image_urls, 5 video_urls and 5 audio_urls in a single request
  • Cite references in the prompt: Refer to them by position as Image 1, Video 1, Audio 1
  • Durations from 2 to 30 seconds (default 5)
  • Three resolutions: 480P, 720P or 1080P (default)
  • Native audio: generate_audio (default true) adds a synchronized audio track at no extra cost
  • Two tiers: Standard, or Prime for faster delivery
  • Async processing: Webhook notification or polling for task completion

​
Use cases

  • Character consistency: Supply reference images of a subject and cite them as Image 1 so the same character carries across shots
  • Motion and voice transfer: Give a video_urls reference for movement, or an audio_urls reference for voice timbre
  • Product animation: Animate a still with start_image_url, or transition to a final shot with end_image_url
  • Dialogue scenes: Write the lines in the prompt and keep generate_audio on
  • Social media content: Vertical 9:16 clips for TikTok, Instagram Reels and YouTube Shorts

​
Generation modes

Keyframes and references are mutually exclusive. A request that mixes start_image_url or end_image_url with any of image_urls, video_urls or audio_urls is rejected.
Reference videos are billed. Their combined duration is added to the output duration, capped at 15 seconds. A 5-second generation with one 4-second reference video is billed as 9 seconds. Reference images and audios add nothing to the price.

​
API Operations

Generate a video with the endpoint of the tier you want, then track it with the list and status endpoints of that same tier. Generation returns a task ID for async polling or webhook notification.

POST /v1/ai/video/wan-3-0

Create a Wan 3.0 video

POST /v1/ai/video/wan-3-0-prime

Create a Wan 3.0 Prime video

GET /v1/ai/video/wan-3-0

List Wan 3.0 tasks

GET /v1/ai/video/wan-3-0-prime

List Wan 3.0 Prime tasks

GET /v1/ai/video/wan-3-0/{task-id}

Get a Wan 3.0 task by ID

GET /v1/ai/video/wan-3-0-prime/{task-id}

Get a Wan 3.0 Prime task by ID

​
Parameters

​
Input limits

​
Frequently Asked Questions

Wan 3.0 is Alibaba’s unified video model. You send a prompt, optionally with a first frame or with reference images, videos and audios, and receive a task ID immediately. Poll the task endpoint or configure a webhook_url to receive the MP4 URL when processing completes. Each request generates one video.
They are the same model with the same parameters and output. Prime runs on Alibaba’s priority queue: a generation finishes in about 2 minutes instead of about 14, at a higher price per second. Use Prime when latency matters and Wan 3.0 when cost does.
By their position in each list: the first item of image_urls is Image 1, the second Image 2; the first item of video_urls is Video 1; the first item of audio_urls is Audio 1. For example: “Image 1 walks along the beach of Video 1, speaking with the voice of Audio 1”.
No. end_image_url requires start_image_url: a request with only a last frame is rejected with a 400 error and is not charged.
The price is per second of the selected resolution. The billed seconds are duration plus the combined duration of the video_urls, capped at 15 seconds. Reference images, reference audios and generate_audio do not change the price. See the Pricing page for current rates.
The one you set in aspect_ratio (default 16:9). With start_image_url, the output follows the first frame instead. resolution is the short side of the video: 480P at 16:9 is 854x480.
The request is rejected with a 400 error that names the field and is not charged. Each reference video must last 1 to 15 seconds, all of them 15 seconds combined, and their combined duration plus duration must not exceed 30 seconds.
Yes. Upload it first through the uploads endpoint (POST /v1/ai/uploads/request-url) and pass the returned file id: upl_img_... in the image fields and upl_vid_... in video_urls. audio_urls takes public HTTPS URLs only.
Rate limits depend on your subscription tier. See Rate Limits for current limits.

​
Best practices

  • Describe action and dialogue: Wan 3.0 follows long, multi-part prompts; write what happens and what is said
  • Cite every reference: A reference the prompt never mentions may be ignored
  • Keep reference videos short: They add to the billed seconds, up to 15 seconds
  • Iterate at 480P: Explore with short, low-resolution generations and render the final clip at 1080P
  • Production integration: Use webhook_url instead of polling for scalable applications
  • Error handling: Implement retry logic with exponential backoff for 503 errors
  • WAN 2.7: The previous Wan generation, with separate text, image, reference and edit endpoints
  • MiniMax Hailuo 3: Reference-guided video generation from MiniMax
  • Seedance 2.5 Pro: ByteDance video generation with reference videos