Skip to content

Latest commit

 

History

History
38 lines (31 loc) · 2.42 KB

README.md

File metadata and controls

38 lines (31 loc) · 2.42 KB
Raw

Comfier

This is just a simple Docker container with a few helpful additions to get you up and running with ComfyUI quickly and conveniently. It already has Python, CUDA, and CuDNN installed, so you only need to run it with --gpus all (already configured in the Compose file) and you won't need any CUDA-related dependencies on your host besides the NVIDIA driver.

I've tested this on Ubuntu 24.04, but I see no reason it wouldn't work on other systems including WSL.

Pre-requisites:

  1. NVIDIA driver 🔗
  2. Docker engine 🔗
  3. NVIDIA Container Toolkit 🔗

Instructions:

  1. Create directories to hold your models, input images, and output images;
  2. Modify docker-compose.yml:
    • Change /path/to/models, path/to/input, and /path/to/output to the paths to the directories you created,
    • Change groupid and userid to the user/group that should read/write the files in the above directories (you can get your current group/user with id -u and id -g, respectively);
  3. Download the models you will want to use (good resources include HuggingFace and Civatai):
    • Checkpoints, e.g. SD1.5 or SDXL,
    • LoRAs,
    • Embeddings,
    • Upscalers,
    • etc;
  4. Run docker compose up in the folder containing docker-compose.yml;
  5. Navigate to http://127.0.0.1:8188 in your browser.
  6. You can use ctrl+c to terminate ComfyUI.

Important note about Licences

When you build the image, quite a number of git repos are pulled in, and for each of them their dependencies are installed. During container startup, quite a number of models are downloaded to get you up and running quickly. Each of these packages, dependencies, and models have their own licences. I haven't gone through them one-by-one (yet), but I believe a lot of them are GPLv3, which is a copyleft licence. Some of the models, such as Stable Diffusion 3, also have non-commercial clauses attached.

To-Do:

  • I will add the list of imported repos and models here, along with their licences.
  • I will release a sample workflow that demonstrates how to do a number of common things, e.g. generation, face detailing, segmentation, inpainting, upscaling, etc.