DOCSEARCHbyAlgolia
Submit Docs

SamurAIGPT Generative Media Skills

repository·main·Indexed 26 days ago

https://github.com/samuraigpt/generative-media-skills

A schema-driven multimodal toolset for AI agents (Claude Code, Cursor, Gemini CLI) to generate and edit professional images, videos, and audio. Powered by the muapi-cli, it provides core primitives for file uploads and editing, alongside an expert library featuring tools like Cinema Director, Nano-Banana, and AI Clipping. It supports the Model Context Protocol (MCP) to expose 19 structured tools to compatible agents and includes specialized scripts for lipsync, image enhancement, and music creation.

Tokens
42.8K
Snippets
70
Records
233
Agent score
87%

What's inside generative-media-skills

  1. Overview of Generative Media Skills architecture

    main

    The project is organized into two main layers to optimize for AI agentic workflows:

    Core Primitives (/core)

    Thin wrappers around muapi-cli providing raw API access for:

    • File upload: core/media/
    • Image editing: core/edit/ (prompt-based)
    • Platform management: core/platform/ (setup, auth, and result polling)

    Expert Library (/library)

    High-level skills that translate creative intent into technical directives, including:

    • Cinema Director: Technical film direction and cinematography.
    • Nano-Banana: Reasoning-driven image generation.
    • UI Designer: High-fidelity mobile/web mockups using Atomic Design.
    • Logo Creator: Minimalist vector branding using Geometric Primitives.
    • Seedance 2 (Doubao Video): Cinematic video generation (text-to-video, image-to-video, and video extension with audio-video sync).
    • AI Clipping: Converts long videos into ranked vertical short clips (Shorts/TikTok/Reels) with server-side transcription and face-tracked auto-crop.
    • YouTube Shorts: Platform-aware presets for social media content.

  2. Use the muapi-photo-pack-generator skill

    main

    The muapi-photo-pack-generator skill (v0.1.0) transforms a single reference photo into a collection of themed images while maintaining high facial identity fidelity. It uses 'Identity Lock Prompting' to prevent the model from generating a new face by referencing the image directly rather than describing the person's features (like age or ethnicity).

    Key Capabilities:

    • Identity Preservation: Maintains exact facial identity, eye shape, nose structure, and jawline.
    • Themed Generation: Creates packs for categories like LinkedIn, Tinder, or Cyberpunk.
    • Vision-First Analysis: Analyzes scene attributes (lighting, framing, expression) without re-describing the person's physical traits.
    # Example execution command
scripts/generate-pack.sh \
  --image "./my_face.jpg" \
  --category "LinkedIn" \
  --identity-lock true \
  --num 5

  3. Compatibility and Integration

    main

    The muapi CLI is optimized for several AI development environments and automation workflows:

    • Claude Code: Supports direct terminal execution via tools and MCP server mode.
    • MCP (Model Context Protocol): Provides a full MCP server with typed input/output schemas.
    • AI IDEs: Seamless integration with Gemini CLI, Cursor, and Windsurf as local scripts.
    • CI/CD: Supports scripting via --output-json, --jq filtering, and semantic exit codes.

  4. Use Prompt Templates for Seedance 2.0

    main

    Seedance 2.0 works best with structured prompts. You can use specific templates for different cinematic needs:

    Cinematic Film

    Use [SCENE], [SUBJECT], [ACTION], [CAMERA], and [STYLE] blocks to define high-fidelity shots.

    Product Ad

    Reference existing videos for editing style and use @Image1 as the hero product. Define timing (e.g., 0–3s) for specific actions.

    Short Drama

    Define scenes with specific timestamps (e.g., 0–5s) and include Dialogue blocks for character speech.

    Advertising / Product Motion

    Focus on [SCENE], [ACTION], [CAMERA], and [STYLE] to achieve commercial-grade results.

    [SCENE] Rain-soaked cyberpunk alley, neon signs reflected on wet cobblestones.
[SUBJECT] A lone figure in a weathered trench coat, face obscured by a wide-brim hat.
[ACTION] Walking slowly, each step splashing neon color into the puddles.
[CAMERA] Low-angle tracking shot, anamorphic lens, slow dolly in. Rack focus to face.
[STYLE] Denis Villeneuve aesthetic, high contrast, desaturated blues and magentas. 24fps.

  5. Set up muapi as an MCP Server

    main

    You can run muapi as a Model Context Protocol (MCP) server to allow MCP-compatible agents (like Claude Desktop or Cursor) to call generation tools directly without using shell scripts. This exposes 19 structured tools including image/video generation, audio creation, and enhancement tools.

    muapi mcp serve

    Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json):

    {
  "mcpServers": {
    "muapi": {
      "command": "muapi",
      "args": ["mcp", "serve"],
      "env": { "MUAPI_API_KEY": "your-key-here" }
    }
  }
}

  8. Use the Floor Plan Rendering skill

    main

    The muapi-floor-plan-rendering skill allows for designing a 2D floor plan and converting it into a realistic, high-quality 3D architectural rendering. It operates in two distinct phases: 2D design and 3D conversion.

    Required Inputs

    • floor_plan_description (text, required): A description of the floor plan (e.g., "a modern 2-bedroom apartment with a balcony and open kitchen").

    Optional Inputs

    • base_plan_image (image_url): An optional 2D floor plan image to use as a starting point.

  9. Best practices for logo generation

    main

    When generating logos using this skill, follow these guidelines for better results:

    • Text Accuracy: Always use ideogram-v3-t2i for logos containing text. If the brand name is misspelled, retry the generation by bolding the exact spelling in the prompt.
    • Badge/Emblem Styles: If the user requests a badge or emblem, append circular badge layout, contained design to the prompt.
    • Abstract Marks: For logos that do not require text (abstract marks), switch the model to flux-2-pro for more creative symbol generation.
    • Prompt Structure: The internal prompt follows this template: Professional logo for "{{brand_name}}" — {{style}}, {{industry}} brand. Color scheme: {{colors}}. {{background}} background. Clean, scalable, vector-quality logo design. Legible brand name text, balanced proportions. Modern 2025 logo design, no drop shadows, no gradients unless requested, isolated on {{background}} background.

  10. Configure muapi API Key

    main

    To use the CLI, you must authenticate with a MUAPI_API_KEY. You can use the interactive configuration command or pass the key directly via a flag. API keys can be managed at the muapi.ai dashboard.

    # Interactive setup
muapi auth configure

# Or pass directly
muapi auth configure --api-key "YOUR_MUAPI_KEY"

  11. Use the Keyboard Art Maker skill

    main

    The muapi-keyboard-art-maker skill generates artistic top-down photographs of keyboard keycaps arranged to spell out custom text messages.

    Required Input

    • display_text (text): The specific text you want to spell out (e.g., "WORKFLOWS ON VADOO AI").

    Execution Steps

    To generate the image, use the muapi image generate command with the ideogram-v3-t2i model using the following prompt template:

    The photograph captures white keycaps, arranged neatly on a dusty black surface to spell "{{display_text}}". Each keycap displays a crisp black letter, with soft, diffused lighting highlighting the subtle shadows, creating a harmonious contrast between the keycaps and their backdrop, while the top-down shot ensures the phrase is clear and in perfect focus. There should be adequate spacing between words. Cinematic quality, professional studio lighting, 8k resolution.

    Supported Aspect Ratios: 1:1 or 4:3.