Woolverine94/biniou

biniou – a self‑hosted AI Studio

What it is

• biniou is an open‑source, locally‑run web UI that bundles dozens of generative‑AI modules. It lets you generate text, images, audio, video, 3‑D shapes and more, all from models you download from Hugging Face. The UI works on Linux, Windows and (experimentally) macOS and can run without a GPU – it will just be slower.

Key capabilities

• Text: llama‑cpp chatbots, multimodal LLaVA, Whisper speech‑to‑text, NLLB translation, prompt generator.
• Images: Stable Diffusion (SD‑1.5, SD‑2.1, SDXL, SD‑3.5, Flux, etc.), Kandinsky, Latent Consistency, PixArt‑Alpha, ControlNet, in‑painting/out‑painting, IP‑Adapter, MagicMix, face‑swap (InsightFace), upscaling (Real‑ESRGAN, GFPGAN).
• Audio: MusicGen, MusicGen‑Melody, MusicLDM, AudioGen, Harmonai, Bark (text‑to‑speech).
• Video: ModelScope txt2vid, Text2Video‑Zero, AnimateDiff, Stable Video Diffusion, video‑to‑video Pix2Pix.
• 3‑D: Shap‑E txt2shape and img2shape.
• Cross‑module workflow – output of one module can be fed directly into another.

Why it matters

• Provides a single, browser‑based control panel for a huge variety of generative models, removing the need to install each tool separately.
• Works offline after the initial model download, which is useful for privacy‑sensitive or air‑gapped environments.
• Low hardware floor (8 GB RAM, CPU‑only) while still supporting CUDA/ROCm acceleration for those with a GPU.

Installation

• One‑click installer scripts for most major Linux distros (OpenSUSE, Rocky/Alma/CentOS/Fedora, Debian/Ubuntu/Pop! OS, Arch‑based CachyOS, OpenMandriva).
• Windows installer (biniou_netinstall.exe or install_win.cmd) that automatically pulls required tools (Git, Python 3.11, Visual Studio Build Tools, ffmpeg, etc.).
• Optional Dockerfile for containerised deployment.
• Prerequisites are installed automatically; the only manual step is cloning the repo and running ./install.sh on Linux.

How to use

1. Open a browser at http://localhost:7860 (default Gradio address).
2. Choose a module from the side panel (e.g., Stable Diffusion).
3. Upload or type your prompt, adjust settings, and hit Generate.
4. Use the Control panel to update, restart, enable authentication, or expose the UI publicly with a single click.

Hardware recommendations

• Minimum: 64‑bit CPU, 8 GB RAM, HDD, internet for model download.
• Recommended: multicore CPU, CUDA‑capable GPU, 16 GB+ RAM, SSD (≈200 GB for a full model set).

Supported model formats

• .gguf for llama‑cpp‑based LLMs (GPT‑OSS, Llama 2/3, Mistral, Mixtral, Qwen, DeepSeek, Gemma, etc.).
• .safetensors for diffusion models (Stable Diffusion, SDXL, Flux, etc.).
• LoRA, textual‑inversion and custom checkpoint files are also accepted.

Community & docs

• Wiki with installation guides, module references, and an updates archive.
• Showroom gallery of generated media.
• Video tutorials on YouTube for Linux, Windows, and overall usage.

License

• Distributed under the MIT license (see LICENSE file).

---

Bottom line: biniou is a comprehensive, locally‑hosted AI studio that brings together a wide spectrum of generative models under a unified, user‑friendly web interface, making it a solid choice for hobbyists, researchers, or anyone who wants to experiment with AI generation without relying on cloud services.