NeuroFusion

NeuroFusion is a modern stable diffusion inspired mobile and web application built with React Native (Expo) that allows users to generate AI-powered images using both text-to-image and image-to-image diffusion techniques. This project was developed during my internship at Innova Solutions.

Features

• Text-to-Image: Generate AI art from written prompts using a custom Stable Diffusion pipeline.
• Image-to-Image: Transform existing images based on new prompts or styles.
• Prompt Queueing: All image generation requests are queued and processed sequentially.
• Cancel Generation: Users can cancel pending or in-progress image generation tasks from their queue.
• Built with Expo for seamless deployment across Android, iOS, and Web.
• FastAPI backend for image processing, queuing, and cancellation logic.
• Integrated CLIP Tokenizer + Custom UNet-based Diffusion Model for high-quality generation.

Tech Stack

• Frontend: React Native (Expo), React Navigation
• Backend: Python, FastAPI, PyTorch, Custom Stable Diffusion
• Database: MySQL / SQLite
• Cloud: Docker

Example Generations

Below are a few example outputs generated using the same prompt. Differences in images arise due to changes in hyperparameters.

App Overview and working

Watch Web Demo (Download/View)

Watch Mobile Demo (Download/View)

Clone the repo

git clone https://github.com/jatinnathh/NeuroFusion.git

cd NeuroFusion

Start Expo frontend

python update_ip.py

netsh advfirewall firewall add rule name="Allow Port 8000" dir=in action=allow protocol=TCP localport=8000

docker compose up --build

Reference papers

• Rombach, R., Blattmann, A., Lorenz, D., Esser, P., & Ommer, B. (2022). High-Resolution Image Synthesis with Latent Diffusion Models.
• Ho, J., Jain, A., & Abbeel, P. (2020). Denoising Diffusion Implicit Models.

Built With

• dockerfile
• javascript
• jupyter-notebook
• python
• shell