- Go to AWS.
- Go to S3 and create an S3 bucket.
- Go to SQS and create a FIFO queue.
- Go to your queue settings and select the option 'Content-based deduplication'.
- Create an IAM user and attach the policy s3_sqs_access.json from this repository.
- Create access keys for the user.
-
Go to RunPod.
-
Go to secure cloud and launch an RTX A6000 pod.
-
Select template RunPod Stable Diffusion. Unselect Start Jupyter Notebook.
-
SSH into your pod.
-
Execute these commands:
git clone https://github.com/JoePenna/Dreambooth-Stable-Diffusion wget https://huggingface.co/panopstor/EveryDream/resolve/main/sd_v1-5_vae.ckpt apt install zip -y mkdir Dreambooth-Stable-Diffusion/training_images mv sd_v1-5_vae.ckpt Dreambooth-Stable-Diffusion/model.ckpt git clone https://github.com/djbielejeski/Stable-Diffusion-Regularization-Images-person_ddim.git mkdir -p Dreambooth-Stable-Diffusion/regularization_images/person_ddim mv -v Stable-Diffusion-Regularization-Images-person_ddim/person_ddim/*.* Dreambooth-Stable-Diffusion/regularization_images/person_ddim/ cd Dreambooth-Stable-Diffusion pip install -e . pip install boto3 pip install pytorch-lightning==1.7.6 pip install torchmetrics==0.11.1 pip install -e git+https://github.com/CompVis/taming-transformers.git@master#egg=taming-transformers pip install captionizer -
Download the files execute_pipeline.py, credentials.py, variables.py and prompts.py from this repository.
-
Go to credentials.py and update it with the access keys credentials you created.
-
Go to variables.py and update it with the name of your S3 bucket and the URL of your SQS queue.
-
Execute the file
python execute_pipeline.py
- Clone this repository.
- Install requirements.
- Go to credentials.py and update it with the access keys credentials you created.
- Go to variables.py and update it with the name of your S3 bucket and the URL of your SQS queue.
- Execute main.py.
- Have fun !
- Web app.
- Model training and inference in a serverless service.
- Explore other technologies for face / person generation.