MediaAgent

AI-powered image generation with natural language. Features an interactive CLI with multiple LLM providers and image backends.

Quick Start

# Install
pip install -e .

# Start the CLI
media-agent

Setup

1. Install

python -m venv .venv
source .venv/bin/activate

# Basic install
pip install -e .

# With cloud LLM providers (recommended)
pip install -e ".[all-llm]"

2. Configure an LLM Provider

Choose one:

OpenAI (easiest)

/settings openai.api_key sk-your-key-here
/llm openai

Anthropic

/settings anthropic.api_key sk-ant-your-key-here
/llm anthropic

Ollama (free, local)

# Install Ollama first: https://ollama.ai
ollama pull llama3.2
ollama serve

/llm ollama llama3.2

Local HuggingFace (requires GPU)

media-agent download llm

/llm huggingface

3. Configure Image Generation

Local GPU (default, requires CUDA GPU)

media-agent download image

/image local

Nanobanana Cloud (no GPU needed)

/settings nanobanana.api_key your-key-here
/image nanobanana

4. Generate Images

Just chat naturally:

> Generate a sunset over mountains
> Create a cyberpunk city at night
> A cat wearing a tiny hat

Or use the direct command:

/generate a beautiful forest landscape

CLI Commands

Command Line Usage

# Interactive mode
media-agent

# Generate directly
media-agent generate "a sunset over mountains"

# With options
media-agent generate "a cat" -r 1024x1024 -s 42

# Download models
media-agent download all

# Manage settings
media-agent settings list
media-agent settings openai.api_key sk-xxx

Python API

from media_agent import MediaAgent

# Using configured settings
agent = MediaAgent()
result = agent.run("Generate a sunset over mountains")
print(result)

# Specify providers
agent = MediaAgent(
    llm_provider="openai",
    llm_model="gpt-4o",
    image_provider="local",
)

# Direct generation (skip LLM)
path = agent.generate(
    prompt="A serene Japanese garden",
    resolution="1344x768",
    seed=42,
)

# Cleanup
agent.unload()

Image Resolutions

Settings Location

Settings are stored in ~/.config/media_agent/settings.json. No environment variables needed.

Requirements

• Python >= 3.10
• For local image generation: CUDA GPU with 8GB+ VRAM
• For cloud providers: API key only

License

MIT