🎨 Creative Moodboard Generator

A lightweight, end‑to‑end creative pipeline that transforms a simple text prompt into a fully generated visual moodboard.
Built with Groq Llama 3.1, HuggingFace SDXL, and a modular agent‑style architecture.

This project is intentionally designed so it can later evolve into a general creative agent (stories, branding kits, design briefs, concept exploration, etc.).

---

🚀 Overview

This system takes a user prompt and produces:

1. Concept Extraction — Identify the key creative themes
2. Image Generation — Produce SDXL images for each concept
3. Evaluation — Score how well the images match the concepts
4. UI Output — Display the final moodboard in a clean Streamlit interface

The architecture is modular, extensible, and ready for future expansion.

---

🧩 Architecture

creative-agent/
│
├── agent/
│   ├── pipeline.py        # Main orchestration logic
│   ├── evaluator.py       # LLM-based scoring
│   └── memory.py          # (optional future expansion)
│
├── tools/
│   ├── intent_parser.py   # Extracts creative concepts
│   └── image_generator.py # Wraps SDXL generation
│
├── models/
│   ├── text_model.py      # Groq Llama 3.1 client
│   └── image_model.py     # HuggingFace SDXL client
│
├── ui/
│   └── app.py             # Streamlit interface
│
└── README.md

Each component is isolated and replaceable — making the system easy to extend.

---

🛠️ Tech Stack

• Groq Llama 3.1 8B Instant — fast concept extraction + evaluation
• HuggingFace SDXL (Router API) — high‑quality image generation
• Streamlit — simple, interactive UI
• Python 3.12
• dotenv — environment variable management

---

🔧 Setup

1. Clone the repo

git clone https://github.com/<your-username>/creative-agent.git
cd creative-agent

2. Create a virtual environment

python -m venv .venv
.\.venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

4. Add your API keys

Create a .env file:

GROQ_API_KEY=your_key_here
HF_TOKEN=your_key_here

---

▶️ Run the App

From the project root:

streamlit run ui/app.py

---

✨ Features

• Extracts 5 creative concepts from any prompt
• Generates SDXL images for each concept
• Scores image–concept alignment using LLM evaluation
• Displays a clean, scrollable moodboard
• Fully modular — ready to expand into a general creative agent

---

🧭 Roadmap

• Add multi‑mode creative workflows (story, branding, design brief, etc.)
• Add a router model for automatic task classification
• Add memory + iterative refinement
• Add Flux.1 or SD3 as optional image backends
• Add async generation for speed

---

💬 Author

Built by Rahmat Abrar Mohammed — Fullstack & AI Engineer.
Focused on agentic systems, creative tooling, and high‑impact product engineering.