Generate images with AI – Stable Diffusion 2 with Docker

2 Comments / AI, Blog / By

Requirements

Build docker image

First of all, create a new folder to host everything that you would need as a workspace by running mkdir generateImages && cd generateImages. Create the following setup.sh file.

Then execute the above file  bash setup.sh to download the models into generateImages/models directory. We download the main repository inside the docker image.  Create the following Dockerfile in the generateImages directory.

setup.sh file: 
#!/bin/bash

# exit when any command fails
set -e

# intialize git lfs
git lfs install

mkdir models
cd models

# download models
# and remove `.git` folders to preserve some space
git clone --depth=1 https://huggingface.co/stabilityai/stable-diffusion-2-1
rm -rf stable-diffusion-2-1/.git
git clone --depth=1 https://huggingface.co/stabilityai/stable-diffusion-2-depth
rm -rf stable-diffusion-2-depth/.git
git clone --depth=1 https://huggingface.co/stabilityai/stable-diffusion-2-inpainting
rm -rf stable-diffusion-2-inpainting/.git
wget https://github.com/intel-isl/DPT/releases/download/1_0/dpt_hybrid-midas-501f0c75.pt
FROM nvcr.io/nvidia/cuda:11.3.0-devel-ubuntu20.04

RUN apt-get update \
    && DEBIAN_FRONTEND=noninteractive apt-get install -y wget

RUN wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh \
    && chmod +x Miniconda3-latest-Linux-x86_64.sh \
    && bash ./Miniconda3-latest-Linux-x86_64.sh -b -p $HOME/miniconda


RUN echo "eval \"\$(/root/miniconda/bin/conda shell.bash hook)\"">>~/.bashrc

RUN DEBIAN_FRONTEND=noninteractive apt-get install -y git ffmpeg libsm6 libxext6

RUN wget https://github.com/git-lfs/git-lfs/releases/download/v3.3.0/git-lfs-linux-amd64-v3.3.0.tar.gz \
    && tar -xf *.tar.gz \
    && cd git-lfs-3.3.0 \
    && ./install.sh \
    && cd \
    && git lfs install

RUN git lfs install && git clone https://github.com/Stability-AI/stablediffusion.git \
    && cd /stablediffusion \
    && sed -i 's/python=3.8.5/python=3.9/g' environment.yaml \
    && eval "$(/root/miniconda/bin/conda shell.bash hook)" \
    && conda env create -f environment.yaml \
    && echo "conda activate ldm" >>~/.bashrc

RUN eval "$(/root/miniconda/bin/conda shell.bash hook)" \
    && conda activate ldm \
    && export CUDA_HOME=/usr/local/cuda \
    && pip install -U --pre triton \
    && pip install gradio timm \
    && conda install -y xformers -c xformers

WORKDIR /stablediffusion
Run the following command to build the docker image. It may take a while. 
docker build -t my_stable_diffusion_image .

Using stable diffusion

Run the following to start the docker container 
docker run --runtime nvidia --gpus all --rm -it \
    -v `realpath imgs`:/stablediffusion/outputs \
    -v `realpath models`:/models \
    -v $HOME/.cache/huggingface:/root/.cache/huggingface \
    -v $HOME/.cache/torch:/root/.cache/torch \
    --name stabdiff \
    --privileged \
    --network host \
    my_stable_diffusion_image bash

Generate images

You can start generating images with the following command and the images will be created in the imgs directory on the host. 
python scripts/txt2img.py \
    --ckpt /models/stable-diffusion-2-1/v2-1_768-ema-pruned.ckpt \
    --config configs/stable-diffusion/v2-inference-v.yaml \
    --device cuda \
    --n_samples 1 \
    --step 150 \
    --H 1024 --W 1024 \
    --n_iter 5 \
    --seed 14 \
    --prompt "a professional photograph of an astronaut riding a horse"

Image Modification with Stable Diffusion

pip install gradio timm \
	&& mkdir midas_models \
	&& cp /models/dpt_hybrid-midas-501f0c75.pt midas_models/
python scripts/gradio/depth2img.py \
    configs/stable-diffusion/v2-midas-inference.yaml \
    /models/stable-diffusion-2-depth/512-depth-ema.ckpt

Image Inpainting with Stable Diffusion

python scripts/gradio/inpainting.py \
    configs/stable-diffusion/v2-inpainting-inference.yaml \
    /models/stable-diffusion-2-inpainting/512-inpainting-ema.ckpt
# or 
streamlit run scripts/streamlit/inpainting.py \
    -- \
    configs/stable-diffusion/v2-inpainting-inference.yaml \
    /models/stable-diffusion-2-inpainting/512-inpainting-ema.ckpt

Enjoy.

 

2 thoughts on “Generate images with AI – Stable Diffusion 2 with Docker”

  1. Michael Gonzalez

    Looks like the dockerfile is pointing to an older release of nvidia cuda that’s no longer hosted.

    I tried updated to a newer release in the dockerfile, works up until the point where it tries to install pytorch and xformers, then it dies incompatible with the current releases 🙁
    Anything you recommend?

    Thank you,
    Mike

    Reply

Leave a Comment

Your email address will not be published. Required fields are marked *

Contact
Contact