Skip to main content
Model ID
Calling method: async

wan3.0-video API Usage Guide

Overview

wan3.0-video is Wan AI’s Wan 3.0 video model. Unlike the wan2.x line, which splits text-to-video, image-to-video and reference-to-video across separate model IDs, Wan 3.0 is a single unified model: the task type is determined by what you put in the media inputs, not by which endpoint you call. Capabilities:
  • Text-to-video — prompt only
  • Image-to-video — supply first_frame, optionally last_frame
  • Reference-to-video — supply reference_image (up to 10) and/or reference_video (up to 5)
  • Video editing / continuation — supply video
  • Duration up to 30 seconds in a single generation (wan2.x caps at 15)
  • Audio generated by default — every output carries an audio track
  • Adaptive aspect ratio — ratio: "adaptive" derives the output framing from the input media
Requests are submitted to the request-queue API and results are retrieved via polling.

Authentication

All API requests require authentication using an API key. Include your API key in the Authorization header:

Submit Video Generation Request

Base URL

Endpoint

Text-to-video

Image-to-video

Supply first_frame. Use ratio: "adaptive" to let the output framing follow the input image instead of being forced to 16:9.

Reference-to-video

Request Parameters

resolution is case-sensitive. The upstream model accepts "480P", "720P" and "1080P" only. A lowercase value such as "720p" is rejected and the request fails with Input should be '1080P', '720P' or '480P': parameters.resolution. The failure is fast (about one second) and the request is refunded automatically, but it will not silently fall back to a working value.
At least one of prompt or a media input is required. Sending neither fails with prompt and media cannot both be empty.

Response

Check Request Status

Endpoint

Example

Response

Request Status Values

Pricing

Billed per second of successfully generated video, by resolution tier. Failed generations are not billed. A 30-second 1080P generation therefore costs $6.00. Keep this in mind before combining the maximum duration with the highest resolution.

List Your Requests

Endpoint

List Available Models

Endpoint

Tips for Better Results

  1. Detailed prompts: describe motion, camera movement, lighting and atmosphere explicitly
  2. Prompt extension: keep prompt_extend enabled for short prompts to let the model enrich them
  3. Negative prompts: exclude common artifacts such as “blurry, low quality, deformed, extra fingers”
  4. Use adaptive for media-driven jobs: with an input image or video, ratio: "adaptive" preserves the source framing; with a text-only prompt it behaves the same as 16:9
  5. Long-form: the 2-30 second range covers a full shot in one generation, but cost scales linearly with duration
  6. Audio: outputs include a generated audio track by default; supply audio_url when you need the motion to follow specific audio
  7. Uppercase resolution: always send 480P / 720P / 1080P, never lowercase