loyal812 / SDXL-Generative-AI

SDXL t2i i2i - base & refiner & scheduler & docker & cicd & github action & makefile & runpod

1. When use my docker hub image.

docker hub: gluck0101/oridosai-txt2img:latest

https://hub.docker.com/repository/docker/gluck0101/oridosai-txt2img

Did you follow me here?
You can see the results after a while. It may take a while to create the first image. It takes time to download a 12GB model.
Speed will improve from the second creation onwards.
Done!

You can also use postman.

Runpod

port 5000 capacity 75GB

2. Manual installation

2.1. For the first time install docker-compose (NOT docker-compose-plugin)

sudo apt-get update
sudo apt-get install docker-compose

2.2. Remove all stopped containers, all networks not used by at least one container, all dangling images, and all dangling build cache

./docker_cleanup.sh

2.3. Clone repository

git clone https://github.com/oridosai/txt2img-sdxl-p1.git
cd txt2img-sdxl-p1

git pull --all
git merge --strategy=recursive --strategy-option=ours --no-ff origin/dev

2.4. Make sure to change the image name in docker-compose.yml and Makefile to your Docker Hub image.

2.5. Build docker image

make build

2.6. Push the docker image to your docker Hub

make push

2.7. Then follow the steps in section 1.

If you want, run locally run

use localhost:5000

make up

Notes:

num_inference_steps is 10 ~ 30

num_inference_steps: dpmpp_sde_k, ddim recommend 10 ~ 15, unipc, dpmpp_2m_k recommend 20 ~ 30

If you set scheduler_name = "" empty string, it will use base model only