Installation

Environment Management

Since different methods might require a different version of libraries We might introduce different conda envs for methods(hopefully there will not be too many)

imagen: The one env that should be able to run most of the code.

git clone https://github.com/TIGER-AI-Lab/ImagenHub.git
cd ImagenHub
conda env create -f env_cfg/imagen_environment.yml
conda activate imagen
pip install -e .

Verify the installation

import imagen_hub

print(imagen_hub.__version__) # should print a string

Extra Setup for Text-guided Image Generation

Running Dall-E

Put your API key in a file openai.env in the keys folder (or anywhere else, the program will be able to fetch it).
You can also follow the suggested way from openai (setting up OS environment variable).

Extra Setup for Subject-driven Image Editing

Downloading weights into `checkpoints` folder

./download_Subject-Driven_IE_weights.sh

Running DreamEdit

DreamEdit depends on SAM / GroundingDINO with CUDA support. Somehow the installation of SAM/GroundingDINO with GPU support requires some extra work on conda.

# make sure you are in a conda environment.
which nvcc # should be something usr/bin/nvcc, otherwise you might skip next line
conda install -c conda-forge cudatoolkit-dev -y 
which nvcc # Now your nvcc path should look something like /home/<name>/anaconda3/envs/<env_name>/bin/nvcc
export BUILD_WITH_CUDA=True
export CUDA_HOME=$CONDA_PREFIX
pip install git+https://github.com/IDEA-Research/GroundingDINO.git
pip install git+https://github.com/facebookresearch/segment-anything.git

Running BLIP-Diffusion

Use the environment blip_environment.yml. Current lavis is not supported in our default yaml.

conda env create -f env_cfg/blip_environment.yml
conda activate imagen_blip

Extra Setup for Multi-concept Image Composition

Downloading weights into `checkpoints` folder

./download_Multi-Concept_IC_weights.sh

Miscellaneous

Updating the environment

# To update env (If it doesnt work, delete the environment and create again)
conda env update imagen --file env_cfg/imagen_environment.yml  --prune

Reinstalling the environment

# If you need to reinstall through deleting the environment and create again:
conda deactivate
conda remove --name imagen --all
conda env create -f env_cfg/imagen_environment.yml
conda activate imagen