The purpose of this repository is to create ML tools with regards to Stable Diffusion. Currently, latent upscaler from this notebook is implemented. So, you can generate 1024 x 1024 pixel resolution images using Stable Diffusion 1.4 like so:
python generate_main.py --prompt "gaelle seguillon, krenz cushart, hyper realistic, low angle shot, wide lens, atmospheric perspective, golden hour, desaturated, split tone, futuristic, complex machinery, small crowds of people, terraforming, final fantasy xiv, aria the animation, venice" --seed 1214452132124
The current CLI command generate_main.py takes in the a --prompt and --seed argument.
Please check the examples images for more sample images.
I've added some here for your viewing pleasure.
Clone this repository to your local environment.
Build it with this command:
docker build -t kasukanra_ml_tools .
For some reason, I wasn't able to have volume syncrhonization using relative paths. As a result, my docker-compose file is using absolute paths for the volume mounts. Make sure to change these volumes paths on your own local environment.
Mounting /.cache saves you the trouble of having to fetch/download models every time you start up the container.
Once your image has been built, start the docker-compose with this command:
docker-compose -p [your_env] up
Assigning the -p flag (project flag) avoids clashing of networks when multiple instances of the same docker-compose are being run.
My version will be:
docker-compose -p yeo_env up
Once the container is running, go into the container with this sample command:
docker exec -it [your-env]-kasukanra_ml-app-1 bash
docker exec -it yeo_env-kasukanra_ml-app-1 bash
This command will be different for you depending on your -p flag name.
Once you are inside the docker container, activate the virtual environment:
source /venv/bin/activate
Example:
python generate_main.py --prompt "gaelle seguillon, krenz cushart, hyper realistic, low angle shot, wide lens, atmospheric perspective, golden hour, desaturated, split tone, futuristic, complex machinery, small crowds of people, terraforming, final fantasy xiv, aria the animation, venice" --seed 1214452132124