Model name.
Supported models:
Hunyuan.
Hailuo.
Kling.
Vidu.
OS.
GV.

Specific version number of the model. By default, the system uses the supported stable version of the model.

1. Hailuo: [02 and 2.3].
2. Kling: [2.0, 2.1, 2.5, O1, 2.6, 3.0, and 3.0-Omni].
3. Vidu: [q2, q2-pro, q2-turbo, q3-pro, and q3-turbo].
4. GV: [3.1].
5. OS: [2.0].

Scenario for the generated video.
Note: Not all models support scenarios.

1. Kling supports motion control (motion_control).
2. Mingmou supports landscape-to-portrait conversion (land2port).
3. Vidu supports special effect templates (template_effect).

Description of the generated video. (Note: A maximum of 2000 characters is supported.) This parameter is required when no reference image is specified.

Specifies the content you want to prevent the model from generating.
Note: Not all models support this.
For example:
Top lighting and bright colors.
People and animals.
Multiple vehicles and wind.

The default value is False, meaning the model follows instructions strictly. For better results with more nuanced prompts, set this parameter to True to automatically optimize the input prompt and improve generation quality.

Image URL for video generation. The URL must be accessible from the public network.
Note:

1. The recommended image size is no more than 10 MB. Different models have different size limits.
2. Supported image formats: JPEG and PNG.
3. If the OS model is used, the input image dimension should be 1280x720 or 720x1280.

The model will generate a video using the image of this parameter as the ending frame.
Models that support this parameter:

1. GV. If the ending frame image is specified, ImageUrl is required as the starting frame.
2. Kling. Version 2.1 supports starting and ending frames with a resolution of 1080P.
3. Vidu. q2-pro and q2-turbo support starting and ending frames.

Note:

1. The recommended image size is no more than 10 MB. Different models have different limits.
2. Supported image formats: JPEG and PNG.

List of up to 3 asset images, used to describe the images the model should use for video generation.

Model that supports multiple images:

1. GV. If multiple images are specified, ImageUrl and LastImageUrl are unavailable.
2. Vidu supports video generation with multiple reference images. The q2 model accepts 1 to 7 images. The ReferenceType in ImageInfos can be used to specify the subject ID for the input.

Note:

1. The image size cannot exceed 10 MB.
2. Supported image formats: JPEG and PNG.

Only Kling O1 supports reference video information.
This can be used as a feature reference video or a video for editing. The default type is video for editing. You can choose to keep the original sound of the video.

Duration of the generated video.
Note:

1. Kling supports 5 and 10 seconds. Default value: 5 seconds.
2. The std mode of Hailuo supports 6 and 10 seconds, and other modes support 6 seconds. Default value: 6 seconds.
3. Vidu supports 1 to 10 seconds.
4. GV supports 8 seconds. Default value: 8 seconds.
5. OS supports 4, 8, and 12 seconds. Default value: 8 seconds.

Additional parameters required.

COS bucket information for the file result. Note: COS is required and the MPS_QcsRole role needs to be created and authorized.

Special scenario parameters required by the model, formatted as a JSON serialized string.
Example:
{"camera_control":{"type":"simple"}}.

API operator name.

Returned task ID after the task is created successfully.
The task progress and generation results can be obtained by calling the query API.