Video Breakdown
An advanced video analysis skill that provides both high-level quality assessment and detailed shot-by-shot breakdowns. It helps content creators, editors, and marketers to objectively evaluate video quality and deconstruct narrative structures. It helps content creators, editors, and marketers to objectively evaluate video quality and deconstruct narrative structures.
This skill is powered by a custom backend service that leverages state-of-the-art video understanding models from fal.ai.
Core Capabilities
| Capability | Description | Use Case |
|---|
| Quality Critique | Provides a quantitative score (1-10) and qualitative comments for key technical aspects of a video. | Quickly evaluate if user-generated content meets quality standards; compare different versions of a video. |
| Shot Breakdown |
Delivers a detailed breakdown of each shot, including timestamps, visual descriptions, and camera work. | Deeply analyze a competitor's video; deconstruct a film scene for educational purposes; create a highlight reel script. |
How It Works
This skill acts as a client to a dedicated backend service. The agent submits a video URL and an analysis_type to the service, which then queues the job and returns a task_id. The agent can then poll the status of the job until it is completed.
Workflow
- 1. Agent: Calls the
/analyze endpoint with video_url and analysis_type. - Service: Submits the job to the AI model and returns a
task_id. - Agent: Periodically calls the
/status/{task_id} endpoint. - Service: Returns the job status (
queued, in_progress, or completed). - Agent: Once completed, retrieves and presents the JSON result to the user.
Usage
1. Quality Assessment
Goal: Get a technical quality report for a video.
Agent Action:
CODEBLOCK0
Expected Output (after polling):
CODEBLOCK1
2. Shot-by-Shot Analysis (拉片)
Goal: Get a detailed scene breakdown of a video.
Agent Action:
CODEBLOCK2
Expected Output (after polling):
CODEBLOCK3
Backend Service API Reference
- -
POST /analyze: Submits a new analysis job.
-
Body:
{"video_url": "string", "analysis_type": "string"}
- -
GET /status/{task_id}: Checks the status of a job. GET /analysis_types: Lists available analysis_type values.
视频拆解
一项先进的视频分析技能，提供高质量评估和详细的逐镜头拆解。它帮助内容创作者、编辑和营销人员客观评估视频质量并解构叙事结构。
该技能由定制后端服务驱动，利用fal.ai最先进的视频理解模型。
核心能力
| 能力 | 描述 | 使用场景 |
|---|
| 质量评判 | 为视频的关键技术方面提供量化评分（1-10分）和定性评论。 | 快速评估用户生成内容是否符合质量标准；比较视频的不同版本。 |
| 镜头拆解 |
提供每个镜头的详细拆解，包括时间戳、视觉描述和摄影手法。 | 深入分析竞争对手的视频；为教学目的解构电影场景；制作精彩片段脚本。 |
工作原理
该技能作为专用后端服务的客户端。智能体将视频URL和analysistype提交给服务，服务将任务排队并返回taskid。智能体随后可以轮询任务状态直至完成。
工作流程
- 1. 智能体：使用videourl和analysistype调用/analyze端点。
- 服务：将任务提交给AI模型并返回taskid。
- 智能体：定期调用/status/{taskid}端点。
- 服务：返回任务状态（queued、in_progress或completed）。
- 智能体：任务完成后，检索并将JSON结果呈现给用户。
使用方法
1. 质量评估
目标：获取视频的技术质量报告。
智能体操作：
json
{
tool: video-breakdown.analyze,
args: {
video_url: https://example.com/my-video.mp4,
analysistype: qualitycritique
}
}
预期输出（轮询后）：
json
{
resolution: {
score: 9,
comment: 分辨率很高，主体异常清晰...
},
lighting: {
score: 9,
comment: 光线极佳，有效利用了鲜艳的霓虹灯...
},
audio: {
score: 7,
comment: 音频清晰，但存在一些背景噪音...
},
stability: {
score: 10,
comment: 视频表现出卓越的稳定性...
}
}
2. 逐镜头分析（拉片）
目标：获取视频的详细场景拆解。
智能体操作：
json
{
tool: video-breakdown.analyze,
args: {
video_url: https://example.com/scene.mp4,
analysistype: shotbreakdown
}
}
预期输出（轮询后）：
json
[
{
start_time: 00:00,
end_time: 00:04,
description: 一位时尚的年轻女性自信地走向镜头...,
camera_work: 镜头固定，采用中低角度拍摄...
},
{
start_time: 00:04,
end_time: 00:08,
description: 女性面部特写，她微微露出笑容...,
camera_work: 镜头缓慢推近，聚焦于她的表情...
}
]
后端服务API参考
- - POST /analyze：提交新的分析任务。
-
请求体：{video
url: string, analysistype: string}
- - GET /status/{taskid}：检查任务状态。
- GET /analysistypes：列出可用的analysis_type值。
