This repository provides a Dockerized environment for running Stability.AI's Stable Diffusion v2 model, targeting artists and designers with limited coding experience. It simplifies the setup and execution of image generation tasks, including text-to-image, image-to-image, and inpainting, with an integrated web UI.

How It Works

The project leverages Docker and NVIDIA Container Toolkit to create a self-contained, GPU-accelerated environment. It utilizes the diffusers library from Hugging Face to load and run Stable Diffusion models. The architecture supports multi-GPU inference through a "Data Parallel" approach, replicating the model across available GPUs to increase throughput. Users can configure FP16 (half-precision) for reduced memory usage or FP32 (single-precision) for potentially higher accuracy.

Quick Start & Requirements

• Install/Run: docker run --name stable-diffusion --pull=always --gpus all -it -p 7860:7860 nicklucche/stable-diffusion
• Prerequisites: Ubuntu (20.04+) or Windows (WSL/Ubuntu recommended), NVIDIA GPU (>= 6GB VRAM), Docker, NVIDIA Container Toolkit, Hugging Face account (for some models).
• Setup: Initial run downloads model weights (~2.8GB+).
• Docs: https://github.com/NickLucche/stable-diffusion-nvidia-docker

Highlighted Details

• Supports Stable Diffusion v2.0 with text-to-image, img2img, and inpainting.
• Multi-GPU inference via Data Parallelism (-e DEVICES=0,1 or -e DEVICES=all).
• FP16 mode for GPUs with <10GB VRAM.
• Customizable model loading via MODEL_ID environment variable.
• Persistent model weights via Docker volumes (-v /path/to/cache:/root/.cache/huggingface/diffusers).

Maintenance & Community

The project appears to be a personal effort with a TODO list indicating planned features. No specific community channels or active contributor information are listed in the README.

Licensing & Compatibility

The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

Model Parallelism is currently disabled. The project is primarily tested on Ubuntu 20.04 and Windows 10 21H2. Performance may vary based on GPU memory distribution in multi-GPU setups.