Model CenterMaiopc

sd_2.5_special_v1

POST/kyyReactApiServer/v2/model-center/tasks

Asset libraries

Usage note: To avoid task failures caused by image, video, or audio review, upload assets to the asset library before using them in generation requests.

Model information

Field	Value
Model ID (`model`)	`sd_2.5_special_v1`
Model type	Video models
Set the request `model` field to the Model ID shown above. Using the model name will fail to resolve the model.

Authentication

Get Key

All requests must include a Bearer token in the request headers:

cURL

Authorization: Bearer <API_KEY>

Request parameters

Send the following fields in the JSON request body. model is required — use the Model ID from the Model information section above.

promptstringrequired

Prompt

Text prompt input.

Minimum：1

reference_imagesimage[]

Reference images

参考图.

Range：0 – 30

Items：0 - 30

Provide a publicly accessible resource URL.

reference_videosvideo[]

Reference videos

Reference video inputs.

Range：0 – 10

Items：0 - 10

Provide a publicly accessible resource URL.

reference_audiosaudio[]

Reference audios

Reference audio inputs.

Range：0 – 10

Items：0 - 10

Provide a publicly accessible resource URL.

durationintegerrequired

Duration

Generated video duration.

Range：4 – 30

Default：4

aspect_ratiostring

Aspect ratio

Output aspect ratio.

Allowed values：16:9, 9:16, 1:1, 4:3, 3:4, 21:9, adaptive

Default：16:9

resolutionstring

Resolution

Output resolution tier.

Allowed values：720p, 1080p

Default：720p

seedinteger

Seed

Deterministic generation seed.

Default：-1

first_imageimage[]

First frame

First-frame image input.

Range：0 – 1

Items：0 - 1

Provide a publicly accessible resource URL.

last_imageimage[]

Last frame

Last-frame image constraint.

Range：0 – 1

Items：0 - 1

Provide a publicly accessible resource URL.

generate_audioboolean

Generate audio

Whether to generate an audio track.

Default：true

omni_reference_task_typestring

Prompt

可选值： auto：由模型根据输入素材和提示词自动判定任务类型。 reference：参考生视频任务，即基于参考图片、参考视频或参考音频生成新视频。设置 reference 时，ratio 或 duration 无特殊限制。 extend：视频延长任务，即对原视频向前或向后延长。设置 extend 时，content 中必须至少包含一个 reference_video；ratio 必须为 adaptive。

Allowed values：auto, reference, extend

return_last_frameboolean

Prompt

是否要求返回尾帧。查询结果中的 lastFrameUrl 用于读取尾帧地址。

Default：false

toolsarray

Prompt

官方工具配置

Asset and Scenario Limits

Prompt text: Chinese prompts should be no more than 500 characters; English prompts should be no more than 1000 words.
Image input: supports jpeg, png, webp, bmp, tiff, gif, heic, and heif; each image must be under 30 MB, aspect ratio must be within (0.4, 2.5), and width/height must be within (300px, 6000px).
Image count: first-frame scenario supports 1 image; first/last-frame scenario supports 2 images; multimodal reference scenario supports 1-9 images.
Video input: supports mp4 and mov; each video duration must be [2, 15] seconds, up to 3 reference videos are allowed, total video duration must not exceed 15 seconds, each video must be under 50 MB, and frame rate must be [24, 60].
Audio input: supports wav and mp3; each audio duration must be [2, 15] seconds, up to 3 reference audio files are allowed, total audio duration must not exceed 15 seconds, and each audio file must be under 15 MB.
Scenario exclusivity: first-frame, first/last-frame, and multimodal reference scenarios (reference image, video, audio) cannot be mixed. Audio cannot be submitted alone and must include at least 1 reference image or video.
The current system does not support base64 input for image/audio. Use public URLs or the assetId returned by the asset upload API.

This model group supports the Seedance2.0 asset library. API users should call the asset upload API to obtain an assetId, wait for review to pass, then use assetId://{assetId} in media URL fields.

Response contract

Task status is normalized to queued, processing, completed, or failed. Result URLs are returned only when status is completed;

Field	Description
`id`	Task ID used to poll status and results.
`object`	Resource type, usually `video` for this model.
`created`	Task creation time as a Unix timestamp in seconds.
`model`	Model name used for this request.
`status`	`queued`, `processing`, `completed`, or `failed`.
`progress`	Task progress, usually returned by the query API.
`result_url`	Generic result URL, returned only after completion.
`video_url`	Typed result URL for this resource, returned only after completion.
`amount`	Credits consumed by this task.
`actualDuration`	Duration used by this task.
`error_message`	Additional result field exposed by model configuration.
`last_frame_url`	Additional result field exposed by model configuration.
`media_url`	Additional result field exposed by model configuration.
`task_id`	Additional result field exposed by model configuration.
`error`	Error message on failure; usually `null` while queued, processing, or completed.

Query task result

Generation tasks run asynchronously. The submit endpoint returns a task id; poll the query endpoint below with that id until status becomes completed or failed.

http

GET https://zcbservice.aizfw.cn/kyyReactApiServer/v2/model-center/tasks/{id}
Authorization: Bearer <API_KEY>

idstringrequired

The task ID returned by the submit endpoint.

When status is completed, the result URL is returned via result_url and video_url. See the Response contract above for the remaining fields.

Poll every 5–10 seconds and stop once status is completed or failed. Result URLs expire, so download and store them promptly.