Skip to content

README.md

Latest commit

 

History

History
232 lines (187 loc) · 8.51 KB

README.md

File metadata and controls

232 lines (187 loc) · 8.51 KB
Raw

English | 中文

Agentic Marketing Content Generation

A multi-agent automated marketing content generation system built on Microsoft Agent Framework and Microsoft Foundry.

From idea to campaign in minutes: Enter a product/topic, get publish-ready marketing materials automatically.

📝 Input: "AI Fitness Coach"
     ↓
🤖 4 Specialized Agents + Real-time Web Research
     ↓
📦 Output: Strategy + Blog + LinkedIn/Instagram/Rednote Posts + Images + TikTok Short Video

Who is this for?

  • Founders & PMs who need marketing content fast
  • Content operators managing multi-platform publishing
  • Developers exploring AI Agent workflows

Generated Content List

Category Content Description
📊 Strategy Marketing Strategy Target audience, pain points, selling points, content framework, tone of voice, brand pillars, keywords
✍️ Copywriting Hero Message One-sentence elevator pitch for the campaign
Blog Article Full long-form article in Markdown format with intro, body, and CTA
Blog Outline Structured outline for the blog content
Social Posts (LinkedIn) Professional tone post with hook, body, and CTA
Social Posts (Instagram) Visual-focused post with hashtags
Social Posts (Rednote) Authentic recommendation-style post
Email Campaign A/B testable subject lines, HTML/plain text body, CTA button, P.S. line
Pain Point Analysis Problem → Solution format analysis
CTA Variations Multiple call-to-action options (direct, curiosity, interactive)
🖼️ Images Image Prompts Detailed prompts for AI image generation (English)
Scene Descriptions Human-readable scene descriptions
Generated Images AI-generated marketing images (PNG files)
🎬 Video Video Script Three-act structure (Problem → Solution → Transformation)
Scene Breakdown Per-scene visuals, voiceover, screen text, duration
SRT Captions Subtitle file in SRT format
Structure Notes High-level video structure summary
Tiktok Short Videos AI-generated video clips (MP4 files)

Architecture Overview

%%{init: {'theme': 'dark', 'themeVariables': { 'fontSize': '16px', 'fontFamily': 'Inter, system-ui, sans-serif', 'primaryColor': '#0ea5e9', 'primaryTextColor': '#f8fafc', 'primaryBorderColor': '#0284c7', 'lineColor': '#38bdf8', 'secondaryColor': '#8b5cf6', 'tertiaryColor': '#06b6d4', 'background': '#0f172a', 'mainBkg': '#1e293b', 'nodeBorder': '#334155', 'clusterBkg': '#1e293b', 'clusterBorder': '#475569', 'titleColor': '#f1f5f9'}}}%%
flowchart TB
    subgraph Workflow[" 🚀 Marketing Workflow "]
        direction TB
        Input(["📝 Topic Input"]):::input --> StrategyPhase
        
        subgraph StrategyPhase[" 🎯 Strategy Phase "]
            SA["Strategy Agent<br/>Single Generation"]:::agent
            
            subgraph DR[" 🔬 DeepResearch Executor "]
                direction LR
                Planner["📋 Planner"]:::research
                Researcher["🔍 Researcher"]:::research
                Analyst["📊 Analyst"]:::research
                Planner --> Researcher --> Analyst
            end
            
            SA -.->|OR| DR
        end
        
        StrategyPhase -->|MarketingStrategy| ContentPhase
        
        subgraph ContentPhase[" ✨ Content Generation "]
            direction TB
            Copy["✍️ Copywriting Agent<br/>Knowledge-based Expert"]:::agent
            Image["🖼️ Image Agent<br/>+ FLUX"]:::agent
            Video["🎬 Video Agent<br/>+ Sora-2"]:::agent
            Copy --> Image --> Video
        end
        
        ContentPhase --> Pack["📦 Packaging Executor"]:::executor
        Pack --> Output(["🎁 CampaignPackage"]):::output
    end
    
    subgraph External[" 🔌 External Function Tools "]
        direction LR
        Tavily[("🔍 Tavily")]:::service
        FLUX[("🎨 FLUX")]:::service
        Sora[("🎥 Sora-2")]:::service
    end
    
    Tavily -.->|web_search| SA
    Tavily -.->|web_search| Researcher
    Tavily -.->|web_search| Copy
    FLUX -.->|generate_image| Image
    Sora -.->|generate_video| Video
    
    classDef agent fill:#0ea5e9,stroke:#0284c7,stroke-width:2px,color:#f8fafc,font-weight:bold
    classDef executor fill:#f97316,stroke:#ea580c,stroke-width:2px,color:#f8fafc,font-weight:bold
    classDef research fill:#06b6d4,stroke:#0891b2,stroke-width:2px,color:#f8fafc,font-weight:bold
    classDef service fill:#8b5cf6,stroke:#7c3aed,stroke-width:2px,color:#f8fafc,font-weight:bold
    classDef input fill:#22c55e,stroke:#16a34a,stroke-width:2px,color:#f8fafc,font-weight:bold
    classDef output fill:#ec4899,stroke:#db2777,stroke-width:2px,color:#f8fafc,font-weight:bold
Loading

Features

  • Modular Agent Design: Four specialized Agents - Strategy, Copywriting, Image, and Video
  • Deep Research Mode: Optional DeepResearchExecutor that performs multi-round web searches for market research
  • AI Content Generation: Integrated FLUX image generation and Sora-2 video generation
  • Knowledge-based Copywriting Style: Copywriting Agent uses an authentic, experience-based writing style
  • Structured Output: All content packaged as CampaignPackage Pydantic model
  • File Persistence: Automatically saved to artifacts/campaigns/<timestamp>/

Quick Start

1. Install Dependencies

pip install -r requirements.txt
pip install agent-framework --pre # or install from source

2. Configure Environment Variables

cp .env.example .env

Required configuration:

# Azure OpenAI (Main Model)
AZURE_OPENAI_ENDPOINT=https://<your-resource>.openai.azure.com/
AZURE_OPENAI_API_KEY=<your-api-key>
AZURE_OPENAI_CHAT_DEPLOYMENT_NAME=gpt-5-mini
AZURE_OPENAI_API_VERSION=2025-04-01-preview

# Tavily Search (Market Research)
Tvly_API_KEY=<your-tavily-key>

Optional configuration (enable AI generation):

# FLUX Image Generation
AZURE_IMAGE_ENDPOINT=https://<your-resource>.openai.azure.com/openai/v1/
AZURE_IMAGE_API_KEY=<your-api-key>
AZURE_IMAGE_DEPLOYMENT_NAME=FLUX.1-Kontext-pro

# Sora-2 Video Generation
AZURE_VIDEO_ENDPOINT=https://<your-resource>.openai.azure.com/openai/v1/videos
AZURE_VIDEO_API_KEY=<your-api-key>
AZURE_VIDEO_DEPLOYMENT_NAME=sora-2

3. Run

# Basic mode
python -m marketing_workflow.cli "AI Fitness Coach"

# Deep research mode: multi-round web search + data-driven strategy
python -m marketing_workflow.cli "AI Fitness Coach" --deep-research

# Full generation: including AI images and videos
python -m marketing_workflow.cli "AI Fitness Coach" --enable-image-gen --enable-video-gen

# Debug mode
python -m marketing_workflow.cli "AI Fitness Coach" --debug

CLI Options

Option Description
--deep-research Enable deep research mode (Planner → Researcher → Analyst)
--enable-image-gen Enable FLUX AI image generation
--enable-video-gen Enable Sora-2 AI video generation
--debug Show Agent execution process
--no-persist Don't save files to disk

Output Structure

artifacts/campaigns/20251201_160510_campaign/
├── manifest.json           # Complete CampaignPackage
├── strategy/
│   ├── strategy.json
│   └── strategy.md
├── copywriting/
│   ├── hero_message.md
│   ├── blog.md
│   └── social_posts.json
├── images/
│   ├── prompts.json
│   └── *.png
└── video/
    ├── video_script.json
    └── *.mp4

Code Usage

from agent_framework.azure import AzureOpenAIChatClient
from marketing_workflow import AgenticMarketingWorkflow, MarketingWorkflowConfig

client = AzureOpenAIChatClient(
    endpoint="https://<resource>.openai.azure.com/",
    deployment_name="gpt-5",
    api_key="<your-key>",
)

workflow = AgenticMarketingWorkflow(
    client,
    config=MarketingWorkflowConfig(
        enable_deep_research=True,
        enable_image_generation=True,
    ),
)

package = await workflow.run("AI Fitness Coach")
print(package.copywriting.hero_message)

Project Structure

marketing_workflow/
├── workflow.py     # Main workflow orchestration
├── agents.py       # Agent definitions and instructions
├── research.py     # Deep research executor
├── schemas.py      # Pydantic data models
├── tools.py        # Tool implementations (Tavily, FLUX, Sora-2)
└── cli.py          # Command line entry point