deep_imagen
deep_imagen copied to clipboard
scripts for running and training imagen-pytorch
Scripts for running imagen-pytorch.
This is currently in development, so things will probably break.
Setup:
python3 -m pip install imagen-pytorch
Running Inference:
python3 imagen.py --imagen yourmodel.pth --tags "1girl, red_hair" --output red_hair.png
Training:
Currently, this is set up to use danbooru-style tags such as:
1girl, blue_dress, super_artist
The dataloader expects a directory with images and tags laid out like this:
dataset/
tags/img1.txt
tags/img2.txt
...
imgs/img1.png
imgs/img2.png
The subdirectories doesn't really matter, only the filenames matter.
To train:
python3 imagen.py --train --source /path/to/dataset --imagen yourmodel.pth
gel_fetch.py
Included is a tool to fetch data from *booru-style websites and creates tag files in the expected format.
Setup:
You will need GelbooruViewer and pybooru to run.
First clone GelbooruViewer into the deep-imagen repo:
cd path/to/deep-imagen/
git clone https://github.com/ArchieMeng/GelbooruViewer
python3 -m pip install git+https://github.com/LuqueDaniel/pybooru
Usage
python3 gel_fetch.py --tags "holo" --txt holo/tags --img holo/imgs --danbooru