Media Generation Workflows

Generated file — do not edit here. Source: JIGGAI/ClawKitchen / docs/MEDIA_GENERATION.md Commit: 91c806e4ff2380a3a4d2e465499c5ce91b15365f Edit: https://github.com/JIGGAI/ClawKitchen/blob/91c806e4ff2380a3a4d2e465499c5ce91b15365f/docs/MEDIA_GENERATION.md

​

Overview

​

Media Node Types

​

media-image

• Marketing visuals and hero images
• Social media graphics
• Product illustrations
• Blog post featured images
• Presentation slides

• Agent Assignment: Team member who executes the generation
• Image Prompt: Text description of desired image
• Media Type: Set to “image” (auto-configured)
• Provider: Auto-discovered from available skills

​

media-video

• Product demo videos
• Marketing commercials
• Educational content
• Social media video posts
• Explainer animations

• Agent Assignment: Team member who executes the generation
• Video Prompt: Description of desired video content
• Media Type: Set to “video” (auto-configured)
• Provider: Auto-discovered from available skills

​

media-audio

• Podcast content
• Voiceovers for videos
• Background music
• Audio advertisements
• Training narrations

• Agent Assignment: Team member who executes the generation
• Audio Prompt: Description of desired audio content
• Media Type: Set to “audio” (auto-configured)
• Provider: Auto-discovered from available skills

​

Skill Auto-Discovery System

1. OpenClaw Skills Directory: ~/.openclaw/skills/
2. Workspace Skills: Team-specific skills in workspace
3. Supported Providers: CellCog, OpenAI, and custom skills

​

How Auto-Discovery Works

• Skill Scanning: System scans for generation scripts in skill directories
• Capability Detection: Identifies image, video, and audio generation scripts
• Provider Selection: Chooses best available provider for each media type
• Environment Setup: Configures API keys and settings from OpenClaw config

​

Supported Providers

​

CellCog Integration

• Multi-modal AI: Support for images, videos, and audio
• High Quality: Production-ready media generation
• Coordination: Multi-agent orchestration for complex content

​

OpenAI Integration

• DALL-E: Image generation via OpenAI API
• Whisper: Audio processing and generation
• API Integration: Direct OpenAI service integration

​

Custom Skills

• Extensible: Add custom generation providers via skill system
• Flexible: Support for specialized or proprietary generation tools
• Configurable: Custom parameters and settings per provider

​

Workflow Integration Patterns

​

Sequential Generation

Draft Content (LLM) → Generate Image → Generate Video → Approval → Publish

​

Parallel Generation

                  → Generate Hero Image
Draft Content →   → Generate Social Video  → Combine Assets → Approval  
                  → Generate Audio Clip

​

Conditional Generation

Analyze Content → Route by Type → Generate Appropriate Media → Review → Publish

​

Iterative Refinement

Initial Generation → Human Review → Refine Prompt → Re-generate → Final Approval

​

Prompt Engineering for Media

​

Image Prompts

• Art style: “photorealistic”, “minimalist”, “hand-drawn”
• Color scheme: “vibrant colors”, “monochromatic”, “sunset palette”
• Composition: “centered subject”, “rule of thirds”, “close-up portrait”

• Main subject: “professional businesswoman”, “modern laptop”, “city skyline”
• Context: “in office setting”, “on wooden desk”, “during golden hour”
• Mood: “confident and approachable”, “sleek and modern”, “warm and inviting”

• Resolution: “high resolution”, “4K quality”, “print ready”
• Format: “landscape orientation”, “square format”, “vertical social media”
• Lighting: “soft natural lighting”, “dramatic shadows”, “bright and airy”

"A modern, clean illustration of a workflow automation dashboard displayed on a sleek laptop screen. The dashboard shows colorful workflow nodes connected by flowing lines. The setting is a bright, contemporary office with soft natural lighting. Style: minimalist vector art with a technology color palette of blues and greens. High resolution, landscape orientation."

​

Video Prompts

• Duration: “30-second video”, “brief 10-second clip”, “2-minute explanation”
• Pacing: “fast-paced montage”, “slow and contemplative”, “energetic presentation”
• Narrative: “product demonstration”, “customer testimonial”, “feature walkthrough”

• Camera work: “smooth camera movements”, “close-up shots”, “wide establishing shots”
• Transitions: “smooth fade transitions”, “quick cuts”, “seamless scene changes”
• Text overlay: “minimal text labels”, “animated titles”, “call-to-action buttons”

• Background music: “upbeat electronic music”, “subtle ambient sounds”, “no background music”
• Voiceover: “professional female narrator”, “energetic male voice”, “conversational tone”
• Sound effects: “subtle interface sounds”, “ambient office noise”, “minimal sound design”

"A 30-second product demo video showing a user creating a workflow in ClawKitchen. Start with a close-up of hands typing, then zoom out to show the visual editor interface. Highlight drag-and-drop functionality with smooth animations. End with the completed workflow running successfully. Style: clean and professional with subtle motion graphics. Include upbeat background music and minimal text overlays showing key features."

​

Audio Prompts

• Gender and age: “professional female voice, mid-30s”, “authoritative male narrator”
• Tone: “warm and friendly”, “confident and professional”, “casual and conversational”
• Accent: “neutral American accent”, “slight British accent”, “international English”

• Pace: “moderate speaking pace”, “slightly faster for energy”, “slow and deliberate”
• Emphasis: “emphasize key benefits”, “casual conversational style”, “educational tone”
• Structure: “introduction, main points, call-to-action”, “storytelling format”

• Quality: “studio quality recording”, “podcast-ready audio”, “broadcast standard”
• Format: “mono voice track”, “stereo with ambient sound”, “voice-only no effects”
• Length: “2-minute narration”, “30-second voiceover”, “brief 15-second intro”

"A 90-second professional voiceover explaining the benefits of workflow automation for small businesses. The narrator should be a confident, approachable female voice with a warm tone. The script should cover time savings, reduced errors, and improved team coordination. Pace should be moderate with natural pauses. Studio quality recording with no background music or effects."

​

Template Variables in Media Nodes

​

From Upstream Nodes

{
  "id": "generate_post_image",
  "kind": "media-image",
  "input": { "from": ["draft_content"] },
  "action": {
    "image_prompt": "Create a featured image for this blog post: {{draft_content.title}}. Style should match this theme: {{draft_content.category}}",
    "mediaType": "image"
  }
}

​

Workflow Variables

{
  "id": "generate_branded_video", 
  "kind": "media-video",
  "action": {
    "video_prompt": "Create a {{workflow.brand_style}} video for {{workflow.campaign_name}} featuring {{workflow.target_audience}}",
    "mediaType": "video"
  }
}

​

Dynamic Content

{
  "id": "personalized_audio",
  "kind": "media-audio", 
  "action": {
    "audio_prompt": "Record a personalized greeting for {{customer.name}} about {{product.name}} with a {{customer.preferred_tone}} tone",
    "mediaType": "audio"
  }
}

​

Managing Generated Assets

​

File Organization

shared-context/workflow-runs/{runId}/
├── run.json
├── node-outputs/
│   ├── 001-generate_image.json
│   └── 002-generate_video.json
└── artifacts/
    ├── hero_image_final.png
    ├── promo_video_v1.mp4
    └── voiceover_final.mp3

​

Asset Metadata

• Original Prompt: Text prompt used for generation
• Provider Used: Which skill/provider generated the asset
• Generation Time: When asset was created
• File Information: Size, format, dimensions/duration
• Quality Metrics: Provider-specific quality scores

​

Asset Delivery

• Reference by Path: Use file path in tool nodes
• Template Variables: Insert asset URLs in content
• Approval Workflows: Include assets in approval requests
• Publishing: Automatically attach to social media posts

​

Troubleshooting Media Generation

​

Common Issues

• Check if media generation skills are installed
• Verify OpenClaw skill directory contains generation scripts
• Confirm API keys are configured in OpenClaw config
• Review skill compatibility with current OpenClaw version

• Increase node timeout settings for complex media
• Check provider service status and rate limits
• Verify network connectivity to generation services
• Consider breaking complex prompts into simpler requests

• Refine prompts with more specific descriptions
• Add technical specifications (resolution, format)
• Include style and mood descriptors
• Test prompts with different providers

• Specify output format and compression in prompts
• Configure provider-specific quality settings
• Consider file size limits for downstream usage
• Implement post-processing compression if needed

​

Debugging Media Workflows

1. Check Provider Availability: Verify generation skills are detected
2. Review Prompt Quality: Test prompts manually with providers
3. Monitor Resource Usage: Watch for memory/disk constraints
4. Validate Configurations: Ensure API keys and settings are correct
5. Test Incrementally: Start with simple prompts and build complexity

​

Optimization Tips

• Use appropriate resolution/quality for intended use
• Cache frequently used prompts and results
• Implement parallel generation for multiple assets
• Consider provider-specific optimization settings

• Monitor API usage and costs across providers
• Implement prompt reuse for similar content
• Use lower-cost providers for draft/preview content
• Set up alerts for unusual usage patterns

• Implement human review steps for critical assets
• Create prompt templates for consistent brand style
• Establish quality guidelines and approval criteria
• Maintain asset libraries for reuse and reference

​

Integration with Publishing Workflows

​

Social Media Publishing

Generate Content → Create Image → Create Video → Schedule Posts → Monitor Performance

​

Blog Publishing

Write Article → Generate Featured Image → Create Social Assets → Publish → Share

​

Marketing Campaigns

Campaign Brief → Generate Assets → A/B Test Variants → Select Winners → Launch Campaign

​

Product Documentation

Feature Description → Generate Screenshots → Create Demo Video → Update Docs → Release