视频画质增强

Summary

The Video Upscaler skill provides professional-grade video quality enhancement by leveraging a powerful, multi-model backend. It intelligently selects the best AI model (Topaz, SeedVR2, etc.) based on the user-defined profile to achieve optimal results, transforming low-resolution or noisy footage into crisp, cinematic-quality video.

This skill abstracts away the complexity of choosing and configuring different AI upscaling models. Instead of dealing with dozens of technical parameters, the user simply chooses a high-level goal, and the skill handles the rest.

Features

- Multi-Model Backend: Dynamically routes requests to the best model for the job (Topaz, SeedVR2, etc.) via a unified API.
Profile-Based Enhancement: Offers a range of pre-configured profiles for common use cases, from standard 2x upscaling to 4K cinematic conversion and 60 FPS frame boosting.
Asynchronous by Design: Handles long-running video processing jobs without blocking the agent.
Simple Interface: Requires only a video URL and a profile name to start.

How It Works

The skill operates in a simple, two-step asynchronous workflow:

1. Submit Job: The agent calls the /upscale endpoint with a video URL and a profile name. The service validates the request, selects the appropriate AI model, and submits the job to the fal.ai backend. It immediately returns a task_id.

2. Poll for Status: The agent uses the task_id to periodically call the /status/{task_id} endpoint. The status will be queued, in_progress, or completed. Once completed, the response will contain the URL of the final, upscaled video.

Available Profiles

Profile Name	Description
INLINECODE8	2x upscale using Topaz Proteus v4. Best all-around quality for live-action footage.
INLINECODE9

Upscale to 4K (2160p) using SeedVR2. Best for cinematic content requiring temporal consistency. | | frame_boost_60fps | 2x upscale + frame interpolation to 60 FPS using Topaz Apollo v8. Best for sports and action. | | ai_video_enhance | 4x upscale using Topaz. Best for AI-generated videos that need resolution boosting. | | web_optimized | Upscale to 1080p with web-optimized H264 output. Best for social media and web publishing. |

End-to-End Example

User Request: "Enhance this video to 4K cinematic quality: [video_url]"

1. Agent -> Skill (Submit Job)

The agent identifies the user's intent and calls the /upscale endpoint with the cinema_4k profile.

CODEBLOCK0

Response:
CODEBLOCK1

2. Agent -> Skill (Poll for Status)

The agent waits and then polls the status endpoint.

CODEBLOCK2

Response (In Progress):
CODEBLOCK3

Response (Completed):
CODEBLOCK4

3. Agent -> User

The agent delivers the final, upscaled video URL to the user.

摘要

视频增强技能通过利用强大的多模型后端，提供专业级的视频质量提升。它能根据用户定义的配置文件智能选择最佳AI模型（Topaz、SeedVR2等），实现最优效果，将低分辨率或带噪视频转化为清晰、电影级的画质。

该技能抽象了选择与配置不同AI增强模型的复杂性。用户无需处理数十个技术参数，只需选择高层次目标，其余工作由技能自动完成。

功能特性

- 多模型后端：通过统一API动态将请求路由至最适合的模型（Topaz、SeedVR2等）。
基于配置文件的增强：提供一系列针对常见场景预配置的配置文件，涵盖标准2倍放大、4K电影级转换及60帧率提升。
异步设计：处理长时间运行的视频任务时不会阻塞代理。
简洁接口：仅需视频URL和配置文件名称即可启动。

工作原理

该技能采用简单的两步异步工作流：

1. 提交任务：代理调用/upscale端点，传入视频URL和配置文件名称。服务验证请求、选择合适AI模型，并将任务提交至fal.ai后端，立即返回task_id。

2. 轮询状态：代理使用taskid定期调用/status/{taskid}端点。状态可能为queued（排队中）、in_progress（处理中）或completed（已完成）。完成后，响应将包含最终增强视频的URL。

可用配置文件

配置文件名称	描述
standardx2	使用Topaz Proteus v4进行2倍放大。实拍素材的最佳全能画质。
cinema4k

端到端示例

用户请求： 将此视频增强至4K电影级画质：[video_url]

1. 代理 -> 技能（提交任务）

代理识别用户意图，使用cinema_4k配置文件调用/upscale端点。

bash
curl -X POST http://backendurl>/upscale \
-H Content-Type: application/json \
-d
videourl: [videourl],
profile: cinema_4k
}

响应：
json
{
task_id: a1b2c3d4-e5f6-7890-1234-567890abcdef,
model_used: fal-ai/seedvr/upscale/video,
profile: cinema_4k
}

2. 代理 -> 技能（轮询状态）

代理等待后轮询状态端点。

bash
curl http://backendurl>/status/a1b2c3d4-e5f6-7890-1234-567890abcdef

响应（处理中）：
json
{
task_id: a1b2c3d4-e5f6-7890-1234-567890abcdef,
status: in_progress,
logs: [Processing frame 100/1200...]
}

响应（已完成）：
json
{
task_id: a1b2c3d4-e5f6-7890-1234-567890abcdef,
status: completed,
result: {
videourl: https://.../upscaledvideo.mp4
}
}

3. 代理 -> 用户

代理将最终增强视频的URL交付给用户。