Wan API

Eight video models behind one key

Four Wan variants — Wan 3.0, Wan 3.0 Prime, Wan 3.0 Pro and Wan 3.0 Prime Pro — sit next to Seedance 2.5, Seedance 2.0, Seedance 2.0 Fast and Seedance 2.0 Mini. GET /v1/models lists what the current key can reach, and each model carries its own documented ID such as wan3.0-video or w3.0-video-prime-pro.

Multimodal input modes

Generation accepts a text prompt alone, a first or last frame, or a reference set of up to 10 images, 5 video clips and 5 audio clips, plus a document or a public web link. Frames and references are separate modes and cannot be mixed in one request; prompt text is capped at 20,000 characters.

Playground with native request preview

Each model page embeds a Playground that builds the flat native request body for the selected model, omits optional fields that were left empty, and checks generation availability before submitting. The same prepaid balance and website session back those runs, so no key has to be pasted into the browser.

Asynchronous task API

Submitting to a model endpoint returns a public task ID immediately; polling reports pending, processing, completed or failed, and a finished response carries a signed Bunny CDN URL that plays or downloads without an Authorization header. POST /v1/responses offers the same generation synchronously, in the background, or over server-sent events.

Rate control through content review

The optional nsfw_checker field selects strict or more permissive review on Wan 3.0 and Wan 3.0 Prime, and each setting has its own per-second rate. Pro variants accept false only. A rejected submission can be retried under the other value with the same public model ID.

Balance and cleanup endpoints

GET /v1/balance returns the spendable account balance plus any remaining limit on the calling key, and DELETE /v1/tasks/{task_id} removes one finished task together with its stored video and last-frame files. Site text is also published for machines at /llms.txt and /llms-full.txt.

• Product and e-commerce teams: animate a still photograph of one product with a slow reveal or camera orbit, then inspect shape, reflections and labels before the clip enters a campaign.
• Script, storyboard and social editors: turn a written scene into a moving preview when no shoot or key artwork exists yet, then judge lighting, camera direction and pace on a short clip before committing to a longer cut.
• Agencies working from briefs: feed a concise document or a public page as input so a written idea becomes a visual starting point for a target audience and one message.
• Engineering teams integrating video: compare Wan and Seedance rates, durations and reference allowances side by side, then keep the winning model behind one key and one balance.
• Reference-driven character work: assign each uploaded asset a role — appearance, movement, sound — so a longer Seedance 2.5 scene keeps distinct characters and locations readable.

Is this an official Alibaba or ByteDance service?

No. Wan API operates independently at wanapi.app and is not affiliated with or endorsed by Alibaba, Wan or ByteDance; those names identify the third-party models exposed through the platform and belong to their owners.

Do I need a subscription to generate video?

There is no monthly video package listed. Access depends on an enabled account with sufficient prepaid USD balance, and each generation is billed as the resolution rate times billable seconds.

What counts as a billable second?

Reference video input seconds plus generated output seconds. Text, image and audio inputs add nothing, and turning generated audio off does not lower the rate.

Can a task be cancelled or retried automatically?

No advertised cancellation, webhook or idempotency-key contract exists, and submissions are not retried or switched to another model automatically. After a timeout or missing task ID, retrieve the existing task first instead of posting again, since a second request could create another paid job.

How long do generated videos stay available?

Completed tasks return a signed CDN URL that needs no key, and files follow the existing storage retention policy. Querying a task does not extend that availability, so copy anything important to your own storage.

Which model should a first test use?

Wan 3.0 covers 480p through 1080p at the lowest rates; Prime is a separate model with different pricing, and the Pro variants are the ones that reach 2K or 4K. Seedance pages list their own duration and reference allowances, so match the deliverable rather than assuming a higher resolution exists on a cheaper model.

Can the same request body be reused across models?

All four Wan models share the 14 native generation fields, but resolution values, model IDs and the nsfw_checker setting differ — Pro variants accept the permissive review value only. Unknown fields are rejected, so check the target model's page before resubmitting.

• Sign in with Google at wanapi.app, open account settings and add funds, since a website account alone does not grant generation access.
• Create a key under API key settings and keep it on the server side; the Playground uses the website session and the same balance instead.
• Pick a model from the overview or GET /v1/models, then read its page for supported resolution values, duration range and reference limits.
• Draft the request in that model's Playground and copy the previewed native body, or build the JSON by hand with the documented model ID.
• Submit the request, store the returned task ID right away, and poll with a five-second starting interval that grows after temporary errors.
• Download or play the signed URL from the completed response, then copy the file to your own storage before retention ends.