Fancy AI

Community

---

Support & Donations

Fancy AI is 100% free and open source. If you'd like to support the ongoing development of local, private AI for mobile:

---

A completely private, offline-first AI world living entirely on your Android device.
Characters who chat, remember, call, post, play, and create — with local AI at the center.

Fancy AI is a fully on-device ecosystem designed to run local language models, image diffusion generators, and speech synthesis directly on Android hardware without requiring an internet connection or cloud subscriptions.

---

Screenshots

---

Key Highlights

• 100% Offline & Private: Your conversations, generated photos, character memories, and voice samples never leave your device.
• On-Device LLMs: Run open-weights language models locally via GGUF (llama.cpp) and LiteRT backends. Cloud API keys (OpenRouter, DeepSeek, Anthropic, OpenAI) remain completely optional.
• Local Diffusion Studio (Aura): Generate images offline using Stable Diffusion 1.5, SDXL, and FLUX.2 Klein models. Features on-device .safetensors to .mnn conversion, LCM acceleration, and ESRGAN upscaling.
• Local Voice & Speech: Talk with characters using on-device Speech-to-Text (Whisper, Zipformer) and realistic Text-to-Speech (Kokoro, Pocket, Supertonic).
• Persistent Character Memory: Instant lexical recall with multilingual keyword ranking, allowing characters to remember past conversations and context with no model download.
• Living Character World: Simulated messaging, multi-character group chats, simulated social networks (Ustagram, Rebbit, Y), and character generation.
• Linux Terminal Subsystem: Integrated PRoot sandboxed Linux environment and terminal emulator inside the app.

---

Architecture & Project Structure

The project is structured into modular Android and C++/Rust libraries:

├── app/          # Jetpack Compose UI, Material 3, navigation, and controllers
├── engine/       # Core LLM inference engines (llama.cpp, LiteRT) and tokenizers
├── image/        # Aura diffusion engine, on-device SafeTensor converter, ESRGAN upscaler
├── voice/        # Local STT (Whisper, Zipformer) and TTS (Kokoro, Pocket, Supertonic)
├── terminal/     # PRoot-based sandboxed Linux subsystem and terminal UI
├── vision/       # Image processing and vision pipeline
└── tools/        # DiT engine patches and model export utilities

---

Hardware Acceleration

• CPU: Highly optimized NEON assembly paths for broad compatibility across 64-bit ARM devices (arm64-v8a).
• GPU (OpenCL / Vulkan): Supported on compatible Adreno and Mali GPUs for image diffusion and neural network layers.
• Qualcomm Snapdragon NPU: Support for Hexagon NPU hardware pipelines on compatible Snapdragon chipsets.

---

Getting Started

Prerequisites

• Android Studio: Ladybug (2024.2.1) or newer
• JDK: Java 21
• Android SDK: Platform API 35 (targetSdk 35, minSdk 28)
• Android NDK: Version 27+ with CMake 3.22+

Clone & Open

git clone https://github.com/Mr-J-369/Fancy-Ai.git
cd Fancy-Ai

Open the project directory in Android Studio. Gradle will sync dependencies and prepare the native C++ toolchain automatically.

Build via Command Line

./gradlew assembleDebug

---

License & Open Source

Fancy AI is licensed under the Apache License, Version 2.0.
You are free to use, modify, and distribute this software under the terms of the Apache 2.0 license.