AI Image Generation in Your Pocket – Fully Offline
Edge Diffusion v1.4 is an Android application that brings the power of Stable Diffusion v1.4 directly to your mobile device - no internet connection required!
Designed and developed as part of my Bachelor’s Final Year Project, this app enables fully offline AI image generation, giving users complete control, privacy and creativity on the go.
Prompt input screen with customizable parameters
Example image generated on-device
- Local Image Generation using the Stable Diffusion pipeline — no cloud dependency.
- Stable Diffusion v1.4 Pipeline – fully implemented in Kotlin and optimized for Android.
- Customizable Parameters for creative control:
- Prompt input for text-to-image generation
- Denoising steps for fine-tuned detail
- Seed control for reproducibility
- Advanced settings (see screenshots below)
- Privacy-First Design: All computations happen on-device, ensuring privacy, security, and zero reliance on external servers.
- Offline by Design: Works without any internet connection — perfect for remote or low-connectivity environments.
- LLM-Powered Prompt Enrichment (Optional): When connected to a local network, the app can leverage a LAN-hosted Language Model to enhance your prompts with richer details (e.g., colors, physical attributes), improving image quality and accuracy.
| Category | Technologies |
|---|---|
| Mobile App | Kotlin, Jetpack Compose |
| AI & Model Deployment | PyTorch, ONNX, Hugging Face |
| LAN Server (Optional) | FastAPI, TinyLlama |
| Other Tools | Android Studio, Gradle, Visual Studio Code |
System Architecture
App Architecture
Below are examples showing how image quality improves as the number of denoising steps increases:
Tip: Typically, 20 steps is a balanced trade-off between quality and speed.
