Visions GUI

An offline python frontend for the QuadVisions Colab Notebook using tkinter.
It offers basic options and interactively displays the generating image. So far PyramidVisions, FourierVisions and CLIP + CPPN are implemented. Image generation code: Jens Goldberg / Aransentin, Colab by Philipuss#4066.


VisionsGUI was tested under Arch Linux and Windows 10 on a NVIDIA RTX 2070 Super. PyramidVisions and FourierVisions require at a card with at least 4GB of VRAM, CLIP + CPPN requires more than 8GB. CPU only is currently not supported.


Python and CUDA are required, the remaining dependencies can be installed via pip. Using a virtual environment like venv is preferred to keep the required pip packages seperate from your existing pip packages.

Arch Linux

  1. Install Python 3 and pip via pacman -S python python-pip
  2. Install the CUDA 11 toolkit via pacman -S cuda
  3. Install the pip requirements: pip install -r requirements.txt


  1. Install Python 3
  2. Install the CUDA 11 toolkit
  3. Install the pip requirements: pip.exe install -r requirements.txt

Other Linux distros

Python 3 should be available as a package for most distros but CUDA 11 might not be. In the latter case, an installer is available on NVIDIA’s site.


To start, run python An internet connection is only required when running a model for the first time. No connection is required after the model has been downloaded.
alt text


The GUI is made up of multiple modules which offer feedback to the user and allow to change a handful of settings.

Image viewer

The image viewer displays the current progress of the image and is periodically updated during generation. The displayed images are also accessible in the images folder.

Progress bar

The progress bar shows the progress of the two stages from left to right.

Prompt bar

The prompt bar consists of a text field and a button labeled GO. A prompt is entered into the text field and the image will be generated as soon as the button is pressed.
The button also serves as the GUI’s state:

Color State
Green The model is ready and is waiting to be started
Yellow The model is currently generating an image
Red An error occured

Settings panel

The settings panel contains a few option for generating an image:

Setting Description
Save every Cycles completed before displaying and saving the next image. Lower values add more overhead.
Rough cycles Cycles during the first stage.
Fine cycles Cycles during the second stage.
Seed If a random or fixed seed is used.
Backend Which model to use

Future stuff

  • Add DirectVisions
  • UI support for multiple text / weight pairs
  • UI support for variable image sizes
  • Compiled executables


View Github