Model facet · API

Wan 3.0 API: 3 Endpoints, 20 References, 30s Max

Wan 3.0 exposes three endpoints: text-to-video, image-to-video and reference-to-video, all called with model identifier alibaba/wan-3.0. Each accepts up to 20 mixed-modality references.

Endpoints and what they accept

EndpointUse whenRequired inputs
text-to-videoPure prompt, no visual inputprompt, aspect_ratio, duration
image-to-videoAnimate a single starting frameprompt, first_frame, aspect_ratio, duration
reference-to-videoHold characters and props consistentprompt, up to 20 mixed-modality references, duration
document-to-videoConvert a slide deck or PDF to a videodocument_reference, prompt, duration

Minimal request

Example
curl -X POST "https://api.oneinfer.ai/v1/ula/generate-video" \
  -H "Authorization: Bearer $ONEINFER_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "alibaba/wan3.0-video",
    "prompt": "A quiet city street at golden hour as camera slowly pans up to the skyline",
    "aspect_ratio": "16:9",
    "resolution": "1080p",
    "duration": 10
  }'

Reference inputs and what they cost

Each reference image, video clip, audio snippet or document adds to the input context but does not change the per-second bill. Use references to anchor character and prop consistency across the clip without inflating the bill.

Ready to test the workflow?

Create account & add credits

Audio in the same pass

Audio is generated alongside the picture in a single asynchronous call — no separate TTS step. Toggle audio on or off per request; off-mode is cheaper because no voice model is invoked.

Cost is driven by duration

Wan 3.0 bills by the second of generated video. Every added second raises cost roughly linearly on every tracked provider; pick the shortest duration that meets the brief before dialing resolution up.

Frequently asked questions

How do I use Wan 3.0 with the API?

Send a POST to /v1/ula/generate-video with model "alibaba/wan3.0-video", a valid API key, and a prompt plus aspect_ratio and duration. Add first_frame, last_frame, or document_reference only when you need them.

How many references can Wan 3.0 accept?

Up to 20 mixed-modality references per request — images, video clips, audio snippets and documents. Seedance 2.5 accepts up to 50; Seedance 2.0 accepts fewer.

Does Wan 3.0 generate audio in the same pass?

Yes. Audio is composed alongside the picture in one asynchronous call — no separate TTS step. Toggle the audio flag per request.

Where is the canonical Wan 3.0 page on OneInfer?

The canonical model page with current pricing, providers, and capabilities is /models/alibaba/wan3.0-video. This page is a focused facet of that entity, not a replacement for it.

How should I treat benchmark or price claims?

Check each claim's provenance label and observed date. Vendor-reported and independently verified numbers are shown as separate evidence classes on this hub.

How do I call Wan 3.0 from the API?

POST to /v1/ula/generate-video with model "alibaba/wan3.0-video", a valid API key, and a prompt plus aspect_ratio and duration. Add first_frame, last_frame, or document_reference only when you need them.

Put Wan 3.0 to work

Fund a controlled evaluation, send a reference frame or document, and measure quality and cost on your own workload.