Overview
Fooocus-cu118 is a Docker build of lllyasviel/Fooocus that targets CUDA 11.8 and cu118 PyTorch wheels. It provides an offline, open-source AI image generation tool based on Gradio, with prompts and images first and minimal manual tuning. The container runs locally and requires only 4GB of Nvidia GPU VRAM, without cloud services, API keys, or subscriptions. The image is built on nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04 and uses torch==2.1.0+cu118 and torchvision==0.16.0+cu118. It is intended for older NVIDIA hardware, including Maxwell and newer GPUs, that newer CUDA 12.x container images no longer support. Startup is via docker compose up -d --build, after which the UI is available at http://localhost:7865. First launch downloads the SDXL base and auxiliary models into the fooocus-data volume. The build includes SDXL base, prompt expansion, inpainting, upscaling, and ControlNet-ready configuration. Models and outputs persist in the named fooocus-data volume across rebuilds. The project is GPL-3.0 licensed.
Pricing
Free
Detailed plans have not been confirmed in our catalog. Check the official website for current limits and billing terms.
Visit WebsitePrices and limits may change. Confirm the currency, billing period, seat minimum and usage caps on the official website.
Use Cases
- Teams use Fooocus-cu118 to run Stable Diffusion-style text-to-image generation on older GPU hardware and CUDA 11.8 drivers.
- Teams use Fooocus-cu118 to run local AI image generation with only 4GB of Nvidia GPU VRAM.
- Teams use Fooocus-cu118 to start the service with one Docker Compose command.
- Teams use Fooocus-cu118 to keep models and outputs across rebuilds.
- Teams use Fooocus-cu118 to expose the UI over the network.
- Teams use Fooocus-cu118 to use prompt expansion, inpainting, and upscaling features.
- Teams use Fooocus-cu118 to install on Unraid using a prebuilt template.
- Teams use Fooocus-cu118 to seed models from a host folder.
Features
Generates images from text prompts
Runs SDXL image generation offline
Supports CUDA 11.8 GPUs
Runs inside a Docker container
Preserves data with a persistent volume
Starts with a single command
Reviews
No reviews yet. Be the first to share your experience!
FAQ
Which GPUs are supported?
Any NVIDIA GPU with a CUDA 11.8+ driver, including Maxwell and newer architectures. Minimal video memory is 4GB.
Why not CUDA 12.x?
Users on older CUDA 11.8 hardware cannot run the official cu121/cu124 images; this build restores compatibility.
Can I expose the UI over the network?
Yes; --listen binds to all interfaces, and --share creates a public Gradio share link, to be used with care.
Are models stored between updates?
Yes; everything is kept in the named fooocus-data volume, so pulling a new image does not remove models or images.
What license applies?
GPL-3.0 — see the LICENSE.