Gemini Omni API 模型解析指南
快速结论
- 定位偏 Flash，优先满足快速生成、低延迟反馈和高频迭代场景。
- 适合概念验证、社媒短视频和批量出片，不必一开始就走重制作流程。
- 使用异步视频任务接口，先提交任务，再通过状态接口轮询结果。
核心能力
- Low-latency output：Built for workflows that need fast feedback and frequent prompt changes.
- Short-form batch generation：Fits short videos, multi-variant testing, and social creative production.
- Unified video task contract：Reuses the shared ToAPIs video task interface for easier production handoff.
适用场景
- When validating concepts, scripts, or motion directions quickly.
- When producing social short-form video or multiple creative variants at scale.
- When low-latency feedback is important for rapid prompt iteration.
不适用场景
- When polished showcase quality and maximum visual stability matter more than speed.
- When you require synchronous immediate responses instead of async task workflows.
运行特性
- Runs as an async video task flow, returning task state first rather than the final video file.
- Best suited to short durations, quick feedback cycles, and repeated parameter tuning.
- Fits server-side queues and production workflows via shared video-task polling patterns.
最小请求示例
{
"model": "gemini_omni",
"prompt": "A cinematic shot of a cat running through neon rain",
"aspect_ratio": "16:9",
"duration": 6,
"resolution": "720P"
}
最小响应示例
{
"id": "video_01JZEXAMPLE",
"object": "generation.task",
"model": "gemini_omni",
"status": "queued",
"created_at": 1779247407
}
关键参数
| 参数 | 类型 | 必填 | 默认值 | 范围 | 说明 |
|---|---|---|---|---|---|
| model | string | 是 | gemini_omni | - | Model identifier. Use the fixed value gemini_omni. |
| prompt | string | 是 | - | - | Main instruction describing the subject, action, scene, and visual style. |
| aspect_ratio | string | 否 | 16:9 | 16:9 | 9:16 |
| duration | integer | 否 | 6 | 6 | 10 |
| resolution | string | 否 | 720P | 720P | Output resolution. Currently only 720P is supported; if omitted, the default value is used. |
常见错误
| HTTP | Code | 触发条件 | 修复建议 | 重试策略 |
|---|---|---|---|---|
| 400 | invalid_request_error | Request payload misses required fields or has invalid field types. | Validate model, mode, and input schema before request. | Retry only after fixing payload; avoid blind retries. |
| 401 | authentication_error | Missing Authorization header or invalid API key. | Verify Bearer token format and API key scope. | Retry after auth fix; rotate key if repeated failures continue. |
| 429 | rate_limit_exceeded | Request rate, concurrency, or current quota hits upstream rate limiting. | Apply exponential backoff first, then review request rate, concurrency, and quota usage. | Use exponential backoff (e.g., 1s/2s/4s) with jitter; if it persists, reduce submission pressure. |
| 500 | internal_error | Transient upstream error or internal processing failure. | Capture request id and trigger retry workflow. | Retry 2-3 times with short delays; escalate if persistent. |
FAQ
- Gemini Omni 更适合什么场景？
更适合概念验证、社媒短视频、批量出片和需要低延迟反馈的高频创意迭代场景。 - 图像视频模型报错：invalid apitype: -1
这类错误通常说明接口走错了。图像和视频模型一般不走 chat 接口，而是按对应文档发起 HTTP 任务请求，并通过任务状态接口轮询结果。排查时建议先看用户的实际请求代码、请求地址和请求体。 - 用户进行生成图片/视频的任务时出现任务失败，但是扣款
先让用户提供任务日志或截图，重点看是否出现了输入或输出 token 统计。如果有这类 token 记录，大概率是用户把图片/视频模型走成了 chat 接口；这不是正确用法。图片和视频模型通常是异步任务接口，需要通过 HTTP 请求先提交任务，再拿到任务 ID 轮询状态，详细以对应文档为准。