wan-3-0, with one difference: no automatic retry. Send model: wan-3-0-basic to POST /api/v1/videos/generations; if a generation fails, the task fails right away and the tokens are refunded, so your own code decides what to do next.wan-3-0, with one difference: no automatic retry. Send model: wan-3-0-basic to POST /api/v1/videos/generations; if a generation fails, the task fails right away and the tokens are refunded, so your own code decides what to do next.
prompt for text-to-video, add image_url for a first frame, and add image_end_url (together with image_url) for a last frame.
wan-3-0 may retry it before reporting; wan-3-0-basic never does. The task ends as failed and the tokens are refunded, which suits pipelines that run their own retry or fallback logic.
duration is any integer from 2 to 30 seconds. generate_audio is on by default and costs nothing extra; turn it off for a silent clip.
model is wan-3-0-basic; prompt describes the video.image_end_url must be used together with image_url. Omit both for text-to-video.480p, 720p or 1080p, default 720p so an omitted value never bills at the top tier. quality is accepted as an alias.adaptive (default, follows the input media), 16:9, 4:3, 1:1, 3:4 or 9:16. There is no 21:9. aspect_ratio is accepted as an alias.generate_audio defaults to true at no extra cost. seed: 0 to 2147483647. custom_callback_url: optional public HTTPS URL.Authorization: Bearer <key>.{"model": "wan-3-0-basic", "prompt": "…", "resolution": "720p", "duration": 5} to /api/v1/videos/generations, adding frame URLs if needed. Save img_uuid.GET /api/v1/jobs/detail or receive the callback. Status moves through pending and processing to success or failed.json[].value within 24 hours. On failure, the tokens are already refunded, so retry, change the prompt, or fall back as your product requires.image_url and image_end_url for previsualization.Same model, parameters and price. The only difference is that wan-3-0-basic never retries a failed generation automatically.
No. This endpoint rejects reference materials; use wan-3-0-basic-refs for reference-driven generation.
Any whole number of seconds from 2 to 30, default 5.
So an omitted value never bills at the most expensive 1080p tier. Set resolution: 1080p explicitly when you need it.
No, generate_audio is free and on by default.
They are refunded automatically as soon as the task fails.