Gradienty was first conceived as a gallery of computer-generated gradients, but early attempts at generating an AI to combine colours produced some truly awful gradients. The AI would need training using two key inputs: user-generated gradients and a concept of what a 'good' gradient is.
Creating a tool for user-generated gradients would be easy enough. The challenge would be understanding what a 'good' gradient is, moreover, doing this without requiring tedious manual labour.
The ideal solution would be a way for visitors to 'like' or 'favourite' existing 'good' gradients and use popular gradients as the basis of a 'good' gradient. I chose to use Tumblr for its flexible API – a gradient could be posted to Tumblr then checked on after a period of time to see users' reactions to it.
Visitors can search through all published gradients in a variety of ways, copy the colour values or complete CSS code, or create a brand new gradient. Tumblr users can like, reblog or comment on the gradient.
A new gradient is published every and, at the time of writing, Gradienty's users and AI have generated a combined total of 50 thousand unique gradients.
Gradienty is full of tiny details like subtle animations. My favourite is the browser favicon that changes to match the open gradient – really useful if you open lots of gradients in different tabs.