imagen
image generator using various Text2Image models
The goal is to compare across a range of different prompt types:
- made up - 'panda riding a bicycle'
- artistic - detailed prompts with styles
- concepts - 'how does a panic attack feel'
- style transfer - xx in style of Warhol
- photos, drawings, illustrations, diagrams
- other types of imaginative prompts
Often using replicate API https://replicate.com/docs/api/getting-started-public
- See list of models in data/configs.py
- Renders get saved into output/renders
- Markdown gallery like this one
Setup
get an API key from replicate and export it https://replicate.com/docs/api/getting-started-public
for example
cp .env.example .env
edit the .env file to add your key then
source .env
# first time: install virtual env:
python3 -m venv venv
# activate venv
source venv/bin/activate
# install pip modules
pip install -r requirements.txt
Running
make run
This will execute the run command inside the Makefile
Depending on what's configured, it should output images into renders/
Have a look at the cli.py to see what commands are available.
Output
Renders get saved into output/renders
Another command makes a markdown gallery like this one
Full link to anchor example:
https://github.com/dcsan/imagen/blob/main/gallery-3.md#octopus-riding-skateboard-22826850
References
settings: https://dazhizhong.gitbook.io/pixray-docs/docs/primary-settings
