How Do I Use the Wan 3.0 API?

Overview

Use the Wan 3.0 API by choosing the correct WaveSpeedAI route, sending a prediction request with supported fields, saving the returned prediction id, and polling the result endpoint for completion. The current WaveSpeedAI example uses the alibaba/wan-3.0/text-to-video route for prompt-based generation.

Begin with the route schema. Current examples include fields such as prompt, resolution, aspect_ratio, duration, enable_prompt_expansion, enable_audio, and seed, but your product should read the exact endpoint before sending production traffic. Do not assume a field from one variant exists on another.

In your backend, treat every generation as a job. Persist request parameters, user id, campaign or project id, prediction id, estimated cost, final status, and output location. Add terminal-state handling for completed, failed, cancelled, timeout, and deleted so support can explain outcomes.

WaveSpeedAI gives developers the callable model route, but the app still owns validation, moderation, retries, storage, and budget controls. Keep Wan 3.0 behind a service boundary so you can switch between text-to-video, image-to-video, reference-to-video, or another model without rewriting the product. That boundary also keeps prompt experiments away from customer-facing guarantees.

Source: WaveSpeedAI Wan 3.0 API collection; Alibaba Cloud Wan3.0 API reference. Source check: 2026-08-29.