图像与视频 API 接入：端点、任务查询与重复提交排查 | Image and Video APIs
区分图片生成、视觉输入和异步视频任务，使用任务 ID 与幂等键管理生成流程。 Integrate image and video generation with correct endpoints and task handling.
中文
先区分三种需求
让模型理解图片属于视觉输入；生成新图片使用图片生成接口；生成视频通常是异步任务。它们不能都塞进普通文本聊天请求。EastX 将这些能力放在同一 API 管理体系下，但调用前仍要选择正确模型和端点。
图片生成
图片生成入口为 POST /v1/images/generations。选择当前支持的图片模型，按其说明设置 prompt、数量与尺寸。响应可能给出生成图片 URL，业务应及时下载并保存所需成果，不把可能过期的返回链接当作永久素材库。不要从某个模型的尺寸选项推断所有图片模型都相同。
视频任务创建与查询
视频入口为 POST /v1/contents/generations/tasks。以下为文档中的请求结构，模型 ID 与可计费分辨率应先查模型详情：
{
"model": "YOUR_VIDEO_MODEL_ID",
"content": [{"type": "text", "text": "A paper plane over a lake."}],
"resolution": "720p",
"duration": 4,
"ratio": "16:9",
"generate_audio": false
}
创建时使用稳定的 Idempotency-Key。保存返回任务 ID，再调用 GET /v1/contents/generations/tasks/{task_id} 查询。queued 与 processing 继续等待；succeeded 后读取 content.video_url；failed 是失败终态。文档建议约每 10 秒查询同一任务。
最容易造成重复费用的错误
页面没更新就重新创建任务，或者网络超时后换一个幂等键重发。相同创建动作重试应复用相同键和请求体。遇到 pending_confirmation 查询原任务并联系支持；billing_pending 则检查余额与原 Key 预算。不要把这些状态当成必须新建任务的信号。
素材与成本
参考图片、视频、音频按模型支持的 content 字段提交，使用可访问的 HTTPS URL 或已授权资产引用；本地文件路径不能直接交给远程 API。费用应结合模型、分辨率、时长和音频选项核对，不能套用文本 Token 的简单比较。
EastX 适合把文本脚本、图片和视频生成整合到同一项目管理中：统一 Key 与消费记录，异步任务保存独立 ID。生产系统应保存输入版本、任务状态和成果位置，方便恢复中断任务。
English
Separate three different needs
Understanding an image is vision input; creating an image uses generation; video generation is typically asynchronous. These are not all ordinary text-chat requests. EastX brings them into one API management system, while each still needs the correct model and endpoint.
Image generation
Use POST /v1/images/generations with a currently supported image model. Configure prompt, count and dimensions according to that model. Responses may provide generated-image URLs; download and store required assets rather than treating a potentially expiring link as a permanent library. Dimensions supported by one model do not establish support for all models.
Create and query video tasks
Use POST /v1/contents/generations/tasks. This follows the documented structure; check the model ID and priced resolutions in model details:
{
"model": "YOUR_VIDEO_MODEL_ID",
"content": [{"type": "text", "text": "A paper plane over a lake."}],
"resolution": "720p",
"duration": 4,
"ratio": "16:9",
"generate_audio": false
}
Send a stable Idempotency-Key when creating the task. Save its ID and query GET /v1/contents/generations/tasks/{task_id}. Continue waiting for queued or processing; read content.video_url after succeeded; failed is terminal. Documentation suggests polling the same task about every 10 seconds.
Avoid duplicate submissions
An unchanged screen or network timeout is not a reason to create another task with a new key. Retry the same creation with the same idempotency key and body. For pending_confirmation, query the original task and contact support. For billing_pending, inspect balance and the original key budget.
Assets and cost
Submit image, video and audio references using supported content fields and accessible HTTPS URLs or authorized asset references. A local file path is not accessible to a remote API. Check model, resolution, duration and audio options when estimating cost; text-token comparisons are insufficient.
EastX helps combine script, image and video production under shared key and usage management. Keep a separate ID for every asynchronous task, and preserve input versions, status and output locations so interrupted production can resume.
资料与更新 / Sources and updates
更新 / Updated: 2026-11-14