This is a little flask website that provides a clean way of demo-ing the latent diffusion model and displaying all the art generated by the convention attendees.
First download the large model:
wget -O models/ldm/text2img-large/model.ckpt https://ommer-lab.com/files/latent-diffusion/nitro/txt2img-f8-large/model.ckpt
Then clone our fork locally. The fork has the tested base docker image that this repo relies on. Then build that container (in that repo's local folder) with the following command:
docker build . -t ldm_base
After you can run docker-compose up --build to run this container locally. You will find the website at localhost:5000. It will output images to the image file here. Run this once with internet access so that HuggingFace's transformers library can cache it's bert model. After that, it can be run without internet access.