vibe-aigc 0.7.1

A New Paradigm for Content Generation via Agentic Orchestration

Vibe AIGC

A New Paradigm for Content Generation via Agentic Orchestration

Based on arXiv:2602.04575

📚 Documentation | 🚀 Quick Start | 📖 API Reference

---

What is Vibe AIGC?

Vibe AIGC bridges the Intent-Execution Gap in AI content generation. Instead of prompt engineering, you provide a Vibe — a high-level representation of your creative intent — and the system automatically decomposes it into executable workflows.

from vibe_aigc import MetaPlanner, Vibe

# Express your intent
vibe = Vibe(
    description="Create a cinematic sci-fi trailer",
    style="dark, atmospheric, Blade Runner aesthetic",
    constraints=["under 60 seconds", "no dialogue"]
)

# Let the Meta-Planner handle the rest
planner = MetaPlanner()
result = await planner.execute(vibe)

Architecture (Paper Section 5)

The implementation follows the paper's three-part architecture:

Component	Purpose	Module
MetaPlanner	Decomposes Vibes into workflows	vibe_aigc.planner
KnowledgeBase	Domain expertise for intent understanding	vibe_aigc.knowledge
ToolRegistry	Atomic tools for content generation	vibe_aigc.tools

from vibe_aigc import MetaPlanner, Vibe, create_knowledge_base, create_default_registry

# The full architecture
kb = create_knowledge_base()  # Film, writing, design, music knowledge
tools = create_default_registry()  # LLM, templates, combine tools

planner = MetaPlanner(knowledge_base=kb, tool_registry=tools)

# Query knowledge for "Hitchcockian suspense" → technical specs
result = kb.query("Hitchcockian suspense")
# Returns: camera techniques, lighting specs, editing patterns

Features

• 🎯 Vibe-based Planning — High-level intent → executable workflows
• 🧠 Domain Knowledge — Built-in expertise for film, writing, design, music
• 🔧 Tool Library — Pluggable tools for actual content generation
• ⚡ Parallel Execution — Independent nodes run concurrently
• 🔄 Adaptive Replanning — Automatic recovery from failures
• 💾 Checkpoint/Resume — Save and restore workflow state
• 📊 Progress Tracking — Real-time callbacks and visualization
• 🎨 Workflow Visualization — ASCII and Mermaid diagrams

LLM Providers

Vibe AIGC supports multiple LLM backends with auto-detection:

Provider	API Key	Model Example
OpenAI	OPENAI_API_KEY	gpt-4
Anthropic	ANTHROPIC_API_KEY	claude-sonnet-4-20250514
Ollama	None (local)	qwen2.5:14b

from vibe_aigc import MetaPlanner, LLMConfig

# Auto-detect (checks env vars, falls back to Ollama)
planner = MetaPlanner()

# Explicit Ollama (no API key needed!)
config = LLMConfig.for_ollama(
    host="http://localhost:11434",  # or your GPU server
    model="qwen2.5:14b"
)
planner = MetaPlanner(llm_config=config)

# Explicit OpenAI
config = LLMConfig.for_openai(model="gpt-4o")
planner = MetaPlanner(llm_config=config)

Installation

pip install vibe-aigc

CLI Usage

# Generate a workflow plan
vibe-aigc plan "Create a blog post about AI" --style "informative" --format ascii

# Execute a vibe
vibe-aigc execute "Design a landing page" --visualize --checkpoint

# Manage checkpoints
vibe-aigc checkpoints --list

Quick Example

import asyncio
from vibe_aigc import Vibe, MetaPlanner

async def main():
    vibe = Vibe(
        description="Write a blog post about AI agents",
        style="informative, engaging",
        constraints=["under 1000 words"]
    )
    
    planner = MetaPlanner()
    result = await planner.execute_with_visualization(vibe)
    
    print(f"Status: {result.get_summary()['status']}")

asyncio.run(main())

Architecture

User Vibe → MetaPlanner → Agentic Pipeline → Execution → Result
     ↑                           ↓
     └──── Feedback Loop ────────┘

Documentation

• Installation Guide
• Quick Start Tutorial
• Core Concepts
• API Reference
• Examples

Contributing

Contributions welcome! See CONTRIBUTING.md for guidelines.

License

MIT — see LICENSE for details.

Citation

@article{vibe-aigc-2025,
  title={Vibe AIGC: A New Paradigm for Content Generation via Agentic Orchestration},
  journal={arXiv preprint arXiv:2602.04575},
  year={2025}
}