simonjoz/runpod-comfy-base repository overview

⁠RunPod Comfy Base

GPU-ready base image for RunPod AI workflows — optimized for performance, extensibility, and compatibility.

⁠Overview

This repository provides a maintained Docker base image built on NVIDIA CUDA and cuDNN with Ubuntu, Python, and key system libraries preinstalled. It’s designed for rapid setup of AI, ML, and ComfyUI-based workflows in containerized environments like RunPod⁠.

⁠Usage

docker pull simonjoz/runpod-comfy-base:0.1.0-cuda12.6.3-ubuntu24.04
docker run --gpus all -it simonjoz/runpod-comfy-base

⁠Core Features

• CUDA 12.6 + cuDNN, based on nvidia/cuda:12.6.3-cudnn-devel-ubuntu24.04
• Python 3.11 with pip and virtualenv preconfigured in /opt/venv
• JupyterLab + notebook extensions and interactive widgets
• Comfy CLI preinstalled for ComfyUI workflow control
• Filebrowser for web-based container file access
• Memory and I/O optimized (e.g., libtcmalloc, no .pyc, pip/huggingface cache tuning)
• Preinstalled media libraries (FFmpeg, libx264, OpenGL, etc.)
• Locale, environment, and cache configurations for seamless AI development

⁠Environment Highlights

• Locale: en_US.UTF-8
• Python path: /opt/venv/bin
• CUDA settings: CUDA_MODULE_LOADING=LAZY, TORCH_CUDNN_V8_API_ENABLED=1
• HuggingFace, pip, virtualenv, and model cache locations optimized under /runpod-volume/.cache

⁠License

MIT