techjarves/Uncensored-Local-Studio

What it solves

Uncensored AI Studio provides a unified, completely offline desktop interface for running multiple AI modalities—image generation, text chat, speech-to-text, and text-to-speech—without requiring cloud subscriptions, API keys, or internet connectivity. It eliminates the complex setup typically associated with local AI by offering a zero-configuration, portable runtime that auto-detects hardware acceleration.

How it works

The studio acts as a GUI wrapper around several high-performance local backends:

• Image Generation: Uses stable-diffusion.cpp to run Stable Diffusion models (SD 1.5, SDXL) via .safetensors, .gguf, or .ckpt files.
• Text Chat: Employs a portable llama.cpp server to run GGUF-format language models.
• Speech-to-Text: Integrates a whisper-cli process powered by whisper.cpp for real-time transcription.
• Text-to-Speech: Uses kokoro-js and the Kokoro-82M ONNX model for natural voice synthesis.

To manage system resources, the image and text engines are mutually exclusive, allowing users to switch between workspaces in the UI. It supports a wide range of hardware acceleration including NVIDIA CUDA, AMD ROCm/Vulkan, Intel OpenVINO (NPU), and Apple Metal.

Who it’s for

Users who want a private, uncensored AI experience on their own hardware across Windows, Linux, and macOS, particularly those who prefer a "plug-and-play" experience over manual environment configuration.

Highlights

• Zero-Install Portability: Includes a self-contained runtime (Node.js and backends) that doesn't change global system settings.
• Hardware Auto-Detection: Automatically configures the best available acceleration (CUDA, ROCm, Vulkan, Metal, or OpenVINO).
• Integrated Model Manager: Allows downloading weights directly from Hugging Face or importing local files via drag-and-drop.
• Real-time Monitoring: Built-in tracking for CPU, RAM, GPU, and VRAM usage within the web interface.
• Cross-Platform: Native support for Windows, Linux, and Apple Silicon Macs.