Wan 3.0 (DashScope)
Wan 3.0 (DashScope)
Section titled “Wan 3.0 (DashScope)”POST /v1/videos
Wan 3.0 video generation through the DashScope-format integration: asynchronous tasks, native synchronized audio, 2–30-second generation (or -1 for model-decided duration), and first/last-frame, reference-image, reference-video, and reference-audio inputs.
Model IDs
Section titled “Model IDs”| Model ID | Positioning |
|---|---|
tk-wan3.0-video | Standard |
tk-wan3.0-video-prime | Faster generation, higher rate |
Both models share the same capabilities and parameters; only speed and price differ.
Supported parameters
Section titled “Supported parameters”Checked against the API request validator on November 10, 2026.
| API field | Type | Required | Values / limit |
|---|---|---|---|
model | string | Yes | tk-wan3.0-video or tk-wan3.0-video-prime |
prompt | string | Yes | Nonempty; at most 6,000 characters |
seconds | integer | No | 2–30, or -1 for model-decided duration; default 5 |
metadata.resolution | string | No | 480P, 720P, 1080P; default 1080P — omitted resolution bills at the 1080P rate |
metadata.ratio | string | No | adaptive (default), 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 |
metadata.media | array | No | {type, url} entries; see reference media below |
metadata.audio | boolean | No | Generated audio track; default true |
metadata.watermark | boolean | No | “AI-generated” watermark; default false |
metadata.seed | integer | No | Random seed |
metadata.prompt_extend | boolean | No | Model rewrites/expands the prompt; default true |
Reference media
Section titled “Reference media”Each metadata.media entry contains type and url:
type | Limit | Notes |
|---|---|---|
first_frame | 1 | — |
last_frame | 1 | Requires first_frame in the same request |
reference_image | 10 | — |
reference_video | 5 | 1–15 s per clip, 15 s total; also used for video editing/extension |
reference_audio | 5 | 1–15 s per clip, 15 s total |
- First/last frames and reference media are mutually exclusive; reference images, videos, and audios combine freely.
- Prompt-referencing: “Image 1”, “Video 1”, “Audio 1” by order within each type.
- URLs must be directly downloadable over public HTTP(S).
fileandlinkmedia types are not supported and are rejected.
Request example
Section titled “Request example”curl https://api.apimart.cc/v1/videos \ -H "Authorization: Bearer $HIPIX_API_KEY" \ -H "Content-Type: application/json" \ --data-raw '{"model": "tk-wan3.0-video", "prompt": "A cat running on a moonlit rooftop, neon lights in the distance", "seconds": 5, "metadata": {"resolution": "720P", "ratio": "16:9", "audio": true}}'Results
Section titled “Results”Save the returned id and poll GET /v1/videos/{id}. On completed, url points to the HiPix content endpoint; you can also request /v1/videos/{id}/content. Check usage logs for the actual charge.
Billing rules
Section titled “Billing rules”- Billed per second at the output resolution rate (see the model square for current rates).
- Reference-video input seconds are billed too, at the same resolution rate: billed seconds = output seconds + reference-video input seconds. The completion response’s resolution (
SR) is the rate actually charged. - First/last frames, reference images, and reference audio are not billed.
- Generated audio,
watermark,seed, andprompt_extenddo not change the price. - Failed tasks are not billed. Content-moderation failures (
DataInspectionFailed) and suspected IP infringement (IPInfringementSuspect) surface as task failures. - Parameter-range violations may fail at submission or as a task failure — both are handled as failures and not billed.
- An unspecified
metadata.resolutionbills at the 1080P rate; always passresolutionexplicitly. duration: -1(model-decided) bills the actual generated seconds; the reservation uses the documented 30-second maximum and is reconciled after completion.