AI Image Generator with Stable Diffusion and CLIP Evaluation

Badaruddin Chachar

Completed work

AI Developer

Hugging Face

Stable Diffusion

Streamlit

Artificial Intelligence

🔠 AI Image Generator with Stable Diffusion + CLIP Evaluation

This Streamlit web app allows you to generate images using various pre-trained Stable Diffusion models and evaluates the generated image against the input prompt using CLIP similarity scoring. The generated images, their prompts, model used, and similarity scores are saved in a SQLite database and displayed in a paginated gallery and sidebar history.

📸 Features

🔥 Choose from multiple Stable Diffusion models (via Hugging Face)

🖼️ Generate high-quality AI images from text prompts

📂 Saves each image, prompt, and model info into a local SQLite database

🧠 Evaluates image-to-prompt similarity using OpenAI's CLIP model

🗝️ Gallery view with pagination and prompt/model/similarity display

📜 Sidebar shows recent image history

🛩️ Clear all history with a single click

🚀 Getting Started

1. Clone the Repository

git clone https://github.com/bchachar/image-generation.git
cd image-generation

2. Set Up a Virtual Environment (recommended)

python3 -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

You need to have PyTorch installed with GPU support (torch.cuda.is_available()) for best performance.

4. Run the App

streamlit run app.py

📆 Requirements

streamlit
transformers
torch
accelerate
scipy
safetensors
diffusers
Pillow
sqlalchemy

💡 How It Works

You enter a prompt and select a Stable Diffusion model.

The app generates an image using the selected model.

The image is evaluated with OpenAI's CLIP model to determine how well it matches the prompt.

The image, prompt, model name, and similarity score are saved in history.db.

The gallery displays the most recent images with prompt, model name, and similarity.

🧠 CLIP Similarity

We use OpenAI’s CLIP model (ViT-B/32) to compute the cosine similarity between the prompt and generated image. This gives a percentage score (0–100) indicating how well the image matches the text prompt.

📁 Folder Structure

.
├── app.py              # Streamlit app
├── requirements.txt    # Python dependencies
├── history.db          # SQLite DB (auto-created)
├── generated_images/   # Folder for saved images
└── README.md           # You're here!

Clear History

Use the "Clear History" button in the sidebar to remove all stored images and data from the database.

📄 License

This project is licensed under the MIT License. See LICENSE file for details.

🌐 Acknowledgements

Hugging Face Diffusers

OpenAI CLIP

Streamlit

Like this project

Completed work

Posted Jun 28, 2025

Created a Streamlit app for AI image generation and evaluation using Stable Diffusion and CLIP.

Likes

Views