A script for generating images that contain hidden text.

Blog post and motivation: https://www.factsmachine.ai/p/hidden-in-plain-sight

This script is designed to run on Modal GPUs. Before running it, please set up your Modal account from your local machine. Please note that Modal is a commercial product

To run the script, clone this repo to your local machine and install the dependencies:

git clone https://github.com/jlowin/aimages
cd aimages
pip install -r requirements.txt

You'll need to create a control image. For best results, use a PNG of white text on a black background, like this one:

Run the script with Modal, replacing the path and prompt appropriately:

modal run sd_controlnet.py \
    --control-image-path path/to/your/control/image \
    --prompt "a prompt to generate the full image" 

This will save an image on your computer.

Note: the first time you run this script, it will download all the model weights and create a Docker image for you. This can take a few minutes, but only needs to be done once. Subsequent runs should start in less than a minute and take only a few seconds to generate an image.

In addition to the control image and prompt, you can provide many options to control image generation: