Overview

The Remix project is a research project that generates visuals using AI technologies.By mixing different image data represented by various layers of reality and virtuality, we create visuals that lie between micro and macro.

For the project, we have started researching the different patterns created randomly in real life. As a starting point, we first looked into the patterns that were created by physically merging materials together. Random and organic patterns could be found which inspired us to look for patterns in other parts of the world.

From there we looked into other types of patterns and searched for contrasting properties from them. Liquid painting works and Google Earth image that was chosen to remix also had a lot of contrasting components. Liquid painting is very random, small in size, and handmade. Google Earth image, on the other hand, is an image on a huge scale and is an image that is already existing. With these two types of images that are similar yet different we used a technique called “textual_inversion” to fine-tune the Stable Diffusion model with a few images. By using the mixed inputs of microscopic “Liquid Painting” and macroscopic “Google Earth View”, we attempted to create visuals with unique textures.

AI Music Generator

In order to explore new possibilities, Maho composed music using AI technology based on the generated visuals.Four different still images were selected from the video, and keywords describing them were generated by the Image to Text AI tool.Here is an example.

Keywords

iridescent texture, fluid energy, ultra detailed content : face, liquified, microscopic, froz, iridiscent, deformations

AI Music Generation Tool

These keywords were then sent to the AI music generation tool to generate a demo of the sound file.Eventually, we used those music as a reference to compose and produce original sounds.