Chrysalis Teaser
Sound reactive workflow for generative image creation
I started experimenting with generative models driven by text inputs for my latest collaboration with sound artist Riccardo Bazzoni - @rrriiikkkyy. This particular output explores the idea of creating hard surfaces such as rocks separated by very soft light challenging the generative artificial model to lean toward some uncommon compositions. Although these models are getting extremely powerful and flexible, I find the most interesting results arise from very meticulous examination of edge cases where the algorithm create artifacts rather than extremely realistic results.
Here's the comparison between the input image and the rendered version. Left -> the map generated by the GAN trained on natural heightmaps, controlled by parameters extracted from the sound. Right -> terrain reproduction rendered in Houdini.#stylegan2 #houdini #chrysalis pic.twitter.com/Wl6jgLYK0F
— Samuel Pietri (@samuelpietri) June 25, 2022
