catspeedcc/sd-webui-forge-docker repository overview

⁠Dockerized version of lllyasviel/stable-diffusion-webui-forge

This is basically a dockerized version of lllyasviel/stable-diffusion-webui-forge which is a stable diffusion webui you use to make AI generated images. Models not included. I needed a dockerized version for my own use, so I made one and now I will give it back to the community :) <3

catspeedcc/sd-webui-forge-docker:latest - currently points to v1.1.2 (recommended)
catspeedcc/sd-webui-forge-docker:v1.1.2 - Menu helper script, fixed spaces, cleaned project root more (2.28 GB)
catspeedcc/sd-webui-forge-docker:v1.1.0 - Small image (2.28 GB)
catspeedcc/sd-webui-forge-docker:v1.0.0 - First release, large image (11.99 GB)

catspeedcc/sd-webui-forge-docker:development - (not supported, parity w/ development branch, if you use it you\'re on your own.)
catspeedcc/sd-webui-forge-docker:bleeding - (not supported, ephemeral, if you use it you\'re on your own.)

Models can be put in sd-webui-forge-docker/models/ directory, organized by type - they will be mounted to the container

Outputs are stored in sd-webui-forge-docker/outputs/txt2img-images/ directory, organized by date

For any DOCKER issues submit a catspeed-cc issue ticket⁠

⁠Development Warning:

Development is completed. v1.0.0 and v1.1.0 are released. I suggest v1.1.0 as the image size is smaller. Use the github instructions for now until I update these ones.

"development" tag is not supported, if you use it you are on your own.

If you find this work useful, there is a star button on both GitHub⁠ and DockerHub⁠ tanks ~ mooleshacat

⁠IMPORTANT cuda notice for v1.1.0 & onwards:

You should be able to use any cuda 12.x version (12.1->12.8) as cuda is backwards and forwards compatible at least within the major version. If you use cuda 12.8 you will need driver 535.13504.05 or higher

⁠v1.1.0 Startup Time Warning:

The first startup time takes a while, it is doing a lot for you in the background. This should become faster on multiple start/stop of the container, but if you docker compose down you will need to wait again on next docker compose up. The container appears to be obliterated when doing so.

As of v1.0.0 you have ability to ./docker-start-containers.sh and ./docker-stop-containers.sh and the docker-init-*.sh and docker-destroy-*.sh scripts (use only one of each) to create and destroy your container.

As of v1.1.0 you have ability to ./docker-reinstall-container-deps.sh which reinstalls the container dependencies while running. It should be noted that if you do this it will be unsupported as the best way to do this is to just ./docker-destroy-multi-gpu.sh and ./docker-init-multi-gpu.sh as it will fetch and reinstall ALL dependencies and sources.

⁠v1.0.0 Large Image Warning:

Holy crap! The image ... YES the image is large. So is this wall of text lol. At least for the image it starts with the fact that we need a full Ubuntu image with cuda12 for this machine learning / AI task. Then you have the original repository being required to fetch other repositories at runtime on launch to function. When I dockerized this everything was "baked into" the image. Unfortunately I do not see any way around this, even if the upstream developers used submodules, they still have to be initialized and "baked into" the image OR fetched at runtime. ML/AI related source repositories and models are very large, due to the nature of the task.

The developers know their own project better than I - and I am a noob. They can integrate it into docker better, and try to cut waste out of the image, but of course all dependencies need to be baked into the image. Otherwise the images will not work, or it would have to fetch them inside the container every time you wanted to docker compose down. It is not the kind of image I would suggest converting to alpine to slim it down, it would be a lot of work and headache. However I found out recently it is not even possible due to cuda being required. I was able to use a cuda/ubuntu base image though and am happy with results of v1.1.0! I am happy to help with anything, but mostly can sit and make my own mess in my repository :)

Do not worry, I have not loaded it with 1000's of models :P

⁠What do you use an image generating AI for?

Currently I use it for nothing but the plan is to use it for readme images if I ever need them, or logos and header images.

In my current setup this is auxiliary to the main task: AI development - "VIBE coding" as the kids call it, man I am getting old... - I plan to use what I learned from this, to segregate my GPU's in my host system, so that each container only accesses one GPU, so I can divide them up. I plan to use LocalAGI/LocalAI federated nodes & codellama13B and Mixtrial 8x7B MOE models which are fine tuned for development, and when my Threadripper build & 2 x SXM4 A100 GPU's arrive, I will get another model to run :)

Not only do I get a docker image, YOU get a docker image too - I feel like Oprah, look under your chair for docker images :P

On top of all that it helps me in my future LocalAGI/LocalAI endeavours! :)

Open Source FTW <3

~ mooleshacat

P.S. I can be more 'loose' in the dockerhub readme, but README.md in GitHub repo is meant to attempt my best to 'blend in' with their original formatting and style. The README.md will likely be reworded in the merge PR if they choose to merge - I think it helps immensely to have docker images, and exposure on dockerhub. They can merge, and push their own images to dockerhub. I already had one hit on GitHub come from DockerHub :)

EOF