Sofia — AI Thumbnail Designer for OpenClaw

Sofia is an AI-powered YouTube thumbnail designer that runs as an OpenClaw agent skill. She generates click-worthy thumbnails using Gemini Flash Image, following a detailed style guide with your photos, brand logos, and proven design patterns.

Give her a video title, and she delivers 3 thumbnail options — composed with your face, relevant brand logos, bold text, and high-contrast layouts optimized for CTR.

What's Included

Prerequisites

• OpenClaw installed and configured
• GEMINI_API_KEY — get a free key at Google AI Studio
• uv package manager
• gh CLI (for cloning only)

Quick Install

git clone https://github.com/benjacord/openclaw-sofia-skill && cd openclaw-sofia-skill && ./install.sh

Manual Installation

1. Clone the repo

   git clone https://github.com/benjacord/openclaw-sofia-skill
   cd openclaw-sofia-skill
   

2. Copy skills to your OpenClaw workspace

   cp -r skills/sofia ~/.openclaw/workspace/skills/
   cp -r skills/thumbnail-agent ~/.openclaw/workspace/skills/
   

3. Copy assets

   cp -r assets/ben-photos ~/.openclaw/workspace/assets/
   cp -r assets/logos ~/.openclaw/workspace/assets/
   

4. Install the generation script

   mkdir -p ~/.openclaw/workspace/skills/nano-banana-pro/scripts
   cp scripts/generate_image.py ~/.openclaw/workspace/skills/nano-banana-pro/scripts/
   

5. Set your API key

   echo "GEMINI_API_KEY=your-key-here" >> ~/.openclaw/.env
   

How to Use Sofia

Once installed, invoke Sofia from your OpenClaw session:

"Sofía, hazme el thumbnail para: How I Automated My Entire YouTube Channel with AI"

Or spawn her as a sub-agent:

sessions_spawn({
    "task": "Generate 3 thumbnail options for the video: 'How I Automated My Entire YouTube Channel with AI'",
    "model": "anthropic/claude-opus-4-6",
    "runTimeoutSeconds": 600
})

Sofia will:

1. Analyze the video title and pick the best composition style
2. Select the right photo of you and relevant brand logos
3. Generate 3 distinct thumbnail options with bold text overlays
4. Deliver them for your approval

Customizing for Your Channel

Replace the photo bank

Delete the contents of assets/ben-photos/ and add your own photos:

• Clear face shots with different expressions (surprised, confident, pointing, etc.)
• Various outfits and backgrounds
• Name them descriptively: you-pointing-glasses.jpg, you-surprised-arms-open.jpg

Update the style guide

Edit skills/thumbnail-agent/style-guide.md to match your brand:

• Your color palette
• Typography preferences
• Layout patterns that work for your audience

Add logos

Drop SVG/PNG files into the appropriate assets/logos/ subfolder:

• ai-tools/ — AI platforms (Claude, ChatGPT, Gemini, etc.)
• automation/ — Automation tools (n8n, Make, Zapier, etc.)
• social/ — Social platforms (YouTube, Instagram, X, etc.)
• dev-tools/ — Developer tools (Cursor, VS Code, GitHub, etc.)
• google/ — Google products (Gmail, Sheets, Calendar, etc.)
• misc/ — Everything else

Folder Structure

openclaw-sofia-skill/
├── README.md
├── install.sh
├── .gitignore
├── skills/
│   ├── sofia/
│   │   └── SKILL.md              # Sofia's personality + rules
│   └── thumbnail-agent/
│       ├── SKILL.md              # Thumbnail generation playbook
│       ├── style-guide.md        # Visual style reference
│       └── best-examples/        # Reference thumbnails
├── scripts/
│   └── generate_image.py         # Gemini Flash Image generator
└── assets/
    ├── ben-photos/               # 50+ reference photos
    └── logos/                    # 50+ brand logos
        ├── ai-tools/
        ├── automation/
        ├── social/
        ├── dev-tools/
        ├── google/
        └── misc/

Troubleshooting

About

Built by Ben Cordero as part of the Imperio Digital AI agent ecosystem, powered by OpenClaw.

Sofia is one of several AI agents that help run a YouTube channel about automation, AI, and building businesses with intelligent tools.

License

MIT — use it, modify it, make it yours.