Click the buttons on the left side to generate an image. The image will be generated right away as soon as you click a button. You can also adjust some settings using the sliders on the top left. Brightness 10 and Pen size 8 are usually optimal.
This is a probabilistic AI image generator. All images you see in this project are generated with pen using code in this project. It uses data from hundreds of other images to predict the color and position of every pixel. Improvements in b1.1: - Increased creativity. I did this by adding more data for more image variety and then by mixing each set of data with each other based on weighted random values. It works well but has some issues with image noise. - Added limited color. It currently can use up to 10 hue values and has full brightness control. Saturation is currently hard coded. This is still experimental and I will focus on color more in the next version. In the next version: - Will focus on data compression so I can fit more image data inside the project. - Add a full color palette. - Improve image clarity and detail. Datasets used: nielsr/CelebA-faces eurecom-ds/celeba Old version: https://scratch.mit.edu/projects/1369041956/
