A London artist named Matt DesLauriers has develop a tool to create a palette from any text prompt, allowing someone to enter a “beautiful sunset” and get a range of colors that match a typical sunset, for example. Or you can be more abstract, find the right color for “a sad and rainy Tuesday”.
DesLauriers has posted his code on GitHub; it requires local Diffuse and Node.JS to be installed. It’s a rough prototype at the moment that requires some technical skill to set up, but it’s also a remarkable example of unexpected graphical innovations that can come from code releases. Open source of powerful image compositing models. Diffusion stable, open-sourced August 22, generates images from a neural network that has been trained on tens of millions of images taken from the Internet. Its ability to draw from a wide range of visual influences transforms the extraction of palette information well.
Other palette examples DesLauriers offers include “Tokyo neon”, which suggests colors from Japan’s vibrant cityscapes, “living coral”, which resembles a coral reef with its deep pink and blue, and “green garden, blue sky”, suggest a shepherd saturating the stage. In a tweet earlier today, DesLauriers prove How different quantization methods (reducing a large number of colors in an image to only a handful representing the image) can produce different color palettes.
This isn’t the first time an artist has used AI to extract color palettes from text. In May, an artist named dribnet published a joint artwork series called “Homage to the Pixel”, inspired by Josef Albers. He also released a online tools which anyone can use to create a six color palette based on text input.
Why use AI to find color palettes? In addition to the novelty element, you can extract the right color from unique sources or abstract feelings like “the day after my last day in high school”, “waste wrapping paper on a sandwich” fast food meat” or “Star Wars and The Lord of the Rings combined.”
The ability to extract color palettes from written prompts seems like something popular art tools might repeat in the future because of choosing groups of colors that work well together being famous is hard. More unexpected applications of image fusion models may be on the way.