🎨 Color Intelligence Engine

A hybrid rule-based + ML-assisted color recommendation system designed to generate high-quality, real-world color palettes and cinematic prompt descriptions for generative AI tools (Mainly for Gemini Ai Pro (Veo 3)).

This project bridges human color theory, real-world semantics, and machine-learned embeddings to produce visually coherent, explainable, and context-aware color recommendations.

---

🚀 What This Project Does

Given:

A base color (e.g. Blue, Purple)

A theme / genre (e.g. Sci-Fi, Mythic, Luxury)

The system:

1. Recommends the best real-world color variants

2. Ranks them using hybrid ML logic

3. Generates a cinematic, model-friendly natural language prompt

4. Displays everything through a modern Flask UI

---

🧠 Why This Project Exists

Generative AI models understand real-world concepts (e.g. Amethyst, Titanium Blue, Velvet Black) far better than abstract color names like “light purple” or “dark blue”.

This engine:

Translates simple user intent → semantically rich color language

Produces outputs that align better with AI visual reasoning

Remains explainable and controllable, not a black box

---

🏗️ System Architecture (High Level)

User Input ↓ Base Color + Theme ↓ Rule-Based Filtering (Ontology) ↓ ML Embedding Similarity (Soft Ranking) ↓ Final Ranking + Explanation ↓ Prompt Assembly ↓ Flask UI Output

---

📊 Datasets Overview

Core (The basic database is manually designed / The choice varies)

Dataset Purpose

base_colours.json User-facing base color anchors real_world_colors.json Grounded real-world color definitions base_realworld_scores.csv Base ↔ real-world compatibility scores enhancements.json Style & texture transformations theme_enhancement_weights.json Theme-context logic

Derived (ML-Generated, Reproducible)

Dataset Purpose

Color Embeddings Semantic similarity between colors Enhancement Embeddings Contextual similarity between enhancements

---

🤖 Where Machine Learning Is Used

This project uses ML correctly and intentionally, not everywhere.

✅ ML is used for:

Text embeddings using a pretrained transformer

Semantic similarity between colors

Soft ranking via cosine similarity

Contextual alignment between themes and enhancements

❌ ML is NOT used for:

Basic color theory rules

Ontology constraints

Hard business logic

This hybrid approach mirrors real-world production systems.

---

🧩 Recommendation Logic

Final score is a weighted combination of:

Final Score = 0.50 × Base Compatibility 0.30 × Embedding Similarity 0.20 × Theme Alignment

This ensures:

Strong base color consistency

Semantic richness

Context awareness

---

📁 Project Structure

color_intelligence_engine/ │ ├── dataset/ │ ├── base_colours.json │ ├── real_world_colors.json │ ├── base_realworld_scores.csv │ ├── enhancements.json │ └── theme_enhancement_weights.json │ ├── engine/ │ ├── init.py │ ├── recommender.py │ └── prompt_builder.py │ ├── web/ │ ├── app.py │ ├── templates/index.html │ └── static/style.css │ ├── scripts/ │ ├── generate_csv.py │ ├── generate_clr_emb.py │ └── generate_enh_emb.py │ ├── requirements.txt ├── README.md └── .gitignore

---

⚙️ Installation & Setup

1️⃣ Clone Repository

git clone https://github.com/osctoss/color_intelligence_engine cd color_intelligence_engine

2️⃣ Create Virtual Environment

python -m venv venv source venv/bin/activate # macOS / Linux venv\Scripts\activate # Windows

3️⃣ Install Dependencies

pip install -r requirements.txt

---

▶️ Running the Application

python web/app.py

---

🧪 Reproducibility

To regenerate derived datasets:

python scripts/generate_csv.py python scripts/generate_clr_emb.py python scripts/generate_enh_emb.py

---

🧠 Example Output

Input

Base Color: Blue

Theme: Sci-Fi

Output

Titanium Blue

Cobalt Royale

Glacier Blue

Generated Prompt

A sci-fi outfit featuring Titanium Blue tones with electric and metallic accents, high detail, realistic fabric and cinematic lighting.

---

🎯 Future Enhancements

Auto-generate HEX values from dataset

User feedback → learning-to-rank

Public deployment (Render / Railway)

REST API endpoints

Visual enhancement previews

---

🧑‍💻 Author Notes

This project was designed to reflect how real ML systems are built:

Human intelligence + machine intelligence

Explainability over blind training

Reproducibility over static artifacts

---

👤 Author

Manish Patel (Osctoss)