Skip to main content
The video generation flow is straightforward: request the model list, select an available model, and then call the generation endpoint to create a task.
If you do not have an API Key yet, read Get API Key first.
View video generation model pricing: International site.

​
Integration order

  1. Call Video generation models to retrieve the video models available on the current site.
  2. Copy the model name from data[n].name and pass it as model to the generation endpoint.
  3. Select the resolution, aspect ratio, duration, and reference assets from the same model item’s resolutions, aspectRatios, durations, modes, and inputConstraints fields.
  4. Call Generate video with model, prompt, and any optional parameters.
  5. The generation endpoint returns data.id. Use it to call Video task status and retrieve the result.
If you need to upload a reference image, video, or audio file, use Asset upload first to obtain a publicly accessible URL, and then pass that URL in the corresponding reference asset field.

​
Model list and modes

GET /api/ai/models/video/generation returns the currently available video models. Do not hard-code model parameters when creating a generation task. Use the model item’s name as model, and select mode, resolution, ratio, and duration from the values supported by that same model item.

​
modes constants

A model may support only some of the modes below. When calling the generation endpoint, mode must be one of the values in the selected model’s modes array. If mode is omitted, the backend infers it from the supplied reference assets. For API integrations, explicitly passing mode is recommended so you can validate that the model supports the requested input type before sending the request. Pricing is determined by the pricing page, the model list endpoint’s pricing field, and the generation endpoint’s data.points value.

​
Minimal request

First, request the model list:
Then create a video generation task:

​
Query the result

Replace {taskId} with the data.id returned by the generation endpoint. When the task is complete, the response includes data.videoUrl and data.coverUrl. To view previous tasks, call Video task list.

​
Current model support

Updated at 07/10/2026, 16:21:07 (Beijing Time).