README.md

---
title: Prompt Search Engine
emoji: 🐠
colorFrom: pink
colorTo: purple
sdk: docker
pinned: false
short_description: Improve image quality with better prompts!
---

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

# Prompt Search Engine

## Overview

This project implements a prompt search engine for Stable Diffusion models. The search engine allows users to input a prompt and returns the top `n` most similar prompts from a corpus of existing prompts. This helps in generating higher quality images by providing more effective prompts.

The search engine consists of two main components:

- **Prompt Vectorizer**: Converts prompts into numerical vectors using a pre-trained embedding model.
- **Similarity Scorer**: Measures the similarity between the input prompt and existing prompts using cosine similarity.

## Setup Instructions

### Requirements

- Python >= 3.9
- pip

### Installation

1. **Clone the repository**

   ```bash
   git clone <repository-url>
   cd <repository-directory>
   ``` 

2. **Create a virtual environment (optional)**

   ```bash
   python -m venv venv
   source venv/bin/activate
   ```

3. **Install dependencies**
   ```bash
   pip install -r requirements.txt
   ```

## Running the `run.py` script
The `run.py` script allows you to run the prompt search engine from the command line.
### Usage
   ```bash
   python run.py --query "Your query prompt here" --n 5 --model "all-MiniLM-L6-v2"
   ```
### Arguments
- `--query`: The query prompt (required).
- `--n`: The number of similar prompts to return (default 5).
- `--model`: The name of the SBERT model to use (default "all-MiniLM-L6-v2").

### Example
`python run.py --query "A cat wearing glasses, sitting at a computer" --n 7`