MediaClaw

Multimodal Agent Platform

Aggregate full-stack AIGC capabilities to quickly build scenario-adapted multimedia generation solutions

English | 简体中文

🚀 Introduction

MediaClaw is an OpenClaw-based multimodal agent platform developed by UnicomAI(YuanJing) team. By aggregating full-category AIGC meta-capabilities including image generation, video creation, speech synthesis, digital human, and post-production effects, it forms a unified and flexible toolset (meta-capability pool) that can be called uniformly and combined flexibly.

We have customized the HMI(Human-Machine Interface) and extended functions and opened it to the public through a unified MediaUI. It is specially designed to support Skill customization for various vertical tasks, helping business teams, developers, and ecological partners quickly build multimedia generation solutions that truly adapt to scenarios, simplifying operations, reducing costs and improving efficiency.

MediaClaw Overall Architecture Diagram

✨ Core Features

• 🎨 Full-stack AIGC Capabilities: Covering full-category multimedia generation capabilities including images, videos, speech, and digital humans
• 🔌 Plugin Architecture: Developed based on OpenClaw ecosystem, seamlessly integrated into existing OpenClaw deployments
• 🎯 Multi-provider Support: Support both YuanJing and SGLang backend providers
• ⚙️ Flexible Configuration: Support configuring different providers and model options by capability dimension
• 🛠️ Out-of-the-box: Provides a complete WebUI interface, ready to use without complex development
• 🔧 Skill Extension: Support custom Skill development to quickly adapt to vertical scenario requirements
• 🎬 Post-processing: Built-in local video processing capabilities such as subtitle burning, green screen matting, video overlay, grading, and audio normalization

📊 Capability Matrix

Feature	Backend Dependency	YuanJing	SGLang	Tool Name
Text to Image	Required	✅	✅	mediaclaw_text_to_image
Image QA	Required	✅	✅	mediaclaw_image_qa
Text to Video	Required	✅ (Wan/Kling)	✅	mediaclaw_text_to_video
Image to Video	Required	✅ (Wan Stylization/Kling Single Image)	✅	mediaclaw_image_to_video
Multiple Images to Video	Required	✅ (Wan Multi-image/Kling First-last Frame)	❌	mediaclaw_images_to_video
Text to Speech	Required	✅	❌	mediaclaw_text_to_speech
Speech Recognition	Required	✅	❌	mediaclaw_speech_recognition
Digital Avatar Video	Required	✅	❌	mediaclaw_digital_avatar
Subtitle Generation	No required (local processing)	N/A	N/A	mediaclaw_build_srt
Subtitle Merge	No required (local processing)	N/A	N/A	mediaclaw_merge_srt
Subtitle Burning	No required (local ffmpeg)	N/A	N/A	mediaclaw_burn_subtitles
Audio Normalization	No required (local ffmpeg)	N/A	N/A	mediaclaw_normalize_audio
Color Grading	No required (local ffmpeg)	N/A	N/A	mediaclaw_apply_grade
Video Overlay	No required (local ffmpeg)	N/A	N/A	mediaclaw_apply_overlay
Green Screen Background Replacement	No required (local ffmpeg)	N/A	N/A	mediaclaw_replace_background
Local Image/Video Processing	No required (local processing)	N/A	N/A	mediaclaw_local_image

📦 Installation Guide

Environment Requirements

• Node.js 22+
• OpenClaw Gateway >= 2026.3.24-beta.2
• ffmpeg is required for using mediaclaw_burn_subtitles / mediaclaw_replace_background

Plugin Installation

# Install MediaClaw plugin
openclaw plugins install ./mediaclaw-plugin --force

# Restart OpenClaw gateway
openclaw gateway restart

WebUI Installation

WebUI is customized based on OpenClaw-Admin:

# Enter OpenClaw-Admin directory
cd OpenClaw-Admin
cp .env.example .env

# Edit openclaw auth token in .env file
OPENCLAW_AUTH_TOKEN=YOUR_AUTH_TOKEN

# Install dependencies
npm install

# Start development server
npm run dev:all

After installation, visit http://localhost:3001/ to use.

⚙️ Configuration Guide

Basic Configuration

Edit the openclaw.json configuration file and add MediaClaw related configuration in the plugins node:

"plugins": {
    "mediaclaw": {
      "enabled": true,
      "config": {
        "providers": {
          "yuanjing": {
            "apiKey": "your-yuanjing-token",
            "baseUrl": "https://maas-api.ai-yuanjing.com"
          },
          "sglang": {
            "baseUrl": "http://sglang-default:30010",
            "apiKey": "default-key"
          }
        },
        "capabilities": {
          "textToVideo": {
            "provider": "yuanjing"
          }
        },
        "defaultProvider": "yuanjing"
      }
    },
  },

Configuration Description:

• YuanJing is used as the default provider by default (defaultProvider: "yuanjing")
• The providers node is the global provider configuration
• capabilities.<name>.provider can specify a provider for each capability individually, overriding the global configuration

Video Model Configuration

YuanJing MaaS platform has integrated Kling services, supporting the selection of Wan or Kling models under the YuanJing provider.

Simplified Configuration (Model Only):

{
  "providers": {
    "yuanjing": { "apiKey": "your-yuanjing-key" }
  },
  "capabilities": {
    "textToVideo": { "videoModel": "kling" },
    "imageToVideo": { "videoModel": "kling" },
    "imagesToVideo": { "videoModel": "kling" }
  }
}

Full Configuration (Provider and Model):

{
  "providers": {
    "yuanjing": { "apiKey": "your-yuanjing-key" }
  },
  "capabilities": {
    "textToVideo": {
      "provider": "yuanjing",
      "videoModel": "kling"
    },
    "imageToVideo": {
      "provider": "yuanjing",
      "videoModel": "kling"
    },
    "imagesToVideo": {
      "provider": "yuanjing",
      "videoModel": "kling"
    }
  }
}

Video Model Options:

• wan - Wan 2.2 Model (Default)
• kling - Kling V3 Model (High-quality Video)

Capability Description:

• image_to_video: Single image to video (supports Wan stylization or Kling single image generation)
• images_to_video: Multiple images/first-last frame to video (supports Wan multi-image or Kling first-last frame generation)

Configuration Parameter Details

Parameter	Description
providers.yuanjing.apiKey	YuanJing API Key (Required)
providers.yuanjing.baseUrl	YuanJing API service address
providers.sglang.baseUrl	SGLang service address
providers.sglang.apiKey	SGLang API Key
providers.sglang.apiPath	API path prefix
capabilities	Capability configuration node, supports: textToImage, textToVideo, imageToVideo, imagesToVideo, imageQA, textToSpeech, digitalAvatar
capabilities.<name>.videoModel	Specify video model under YuanJing provider: wan or kling
defaultProvider	Default provider for capabilities not individually configured
outputDir	Output directory for generated files
videoPollInterval	Video generation polling interval (ms), default 5000
videoMaxWaitTime	Maximum waiting time for video generation (ms), default 300000

🔍 SGLang Vision Configuration

The mediaclaw_image_qa capability uses the OpenAI compatible Vision interface in sglang mode:

• Interface address: POST /chat/completions

Supports two path formats:

• /v1/chat/completions
• /openapi/v1/web_control/chat/completions

Configuration Suggestions:

• If baseUrl already includes /openapi/v1/web_control, set apiPath to an empty string
• If baseUrl is only the host address (e.g., http://127.0.0.1:30010), set apiPath to /v1

🛠️ Built-in Skills

Long Video Generation

• Detailed description: skills/unicom-longvideo/SKILL.md

demo_longvide.mp4

Product Poster Generation

Generate product promotional posters with MediaClaw from a product brief, then review each result against a fixed marketing scorecard. The skill supports an iterative generate-review-improve loop for campaign visuals instead of a one-shot image workflow.

• Detailed description: mediaclaw-plugin/skills/unicom-product-poster/SKILL.md

Digital Avatar Production

• Detailed description: skills/unicom-digital-avatar/SKILL.md

demo_avatar.mp4

Video cut

• Detailed description: skills/unicom-video-cut/SKILL.md

🙏 Acknowledgments

The development of MediaClaw is inseparable from the support of the open source community. We would like to express our special thanks to:

• OpenClaw - Provides a powerful plugin gateway platform and ecological support, which is the operating foundation of MediaClaw
• OpenClaw-Admin - Provides an excellent management interface framework, based on which we have customized and extended AIGC capabilities
• All developers who contribute to open source projects

📄 License

MediaClaw is open sourced under the MIT License. You are free to use, modify and distribute it, but please retain the relevant copyright notice and acknowledgment information.

---

If this project is helpful to you, please give us a ⭐️ Star to support!