Speculative Anthropology.
Collaborators:
GPT-3, Jukebox AI, Latent3Visions: CLIP+Taming, Halluzinator 0.4, Soundcloud, Adobe Audition
Methods:
Music
The music was created in collaboration with GPT-3. I first created a set of song titles in the style of The Smiths. Then using those outputs I again prompted GPT-3 to create lyrics based on those titles. I then used 12 second samples of music from The Smiths and the generated lyrics as prompts for the machine learning model JukeBox AI. This process takes around 3 days to complete. I also did some post production using Adobe Audition and hosted the music on Soundcloud.
Poetry
The poetry was generated through a collaboration with GPT-3. The system is very simple and only requires at is most basic a simple prompt such as “A poem about RoboCop by Frank O’Hara.” The model has a series of fine tuning settings that can be adjusted to impact the output of the text also.
Posters
The posters are created in collaboration is a few of machine learning models, primarily Letent3Visions, and Halluzinator. The are very powerful text-to-image models using CLIP from Open AI and a model specific GAN. I provide a series of prompts and keywords to kick of the making and then the GAN works with/against CLIP to create and then identify images and compare them to my prompt for accuracy. Each time attempting to create something that the network thinks satisfies my prompt is some way.
GPT-3, Jukebox AI, Latent3Visions: CLIP+Taming, Halluzinator 0.4, Soundcloud, Adobe Audition
Methods:
Music
The music was created in collaboration with GPT-3. I first created a set of song titles in the style of The Smiths. Then using those outputs I again prompted GPT-3 to create lyrics based on those titles. I then used 12 second samples of music from The Smiths and the generated lyrics as prompts for the machine learning model JukeBox AI. This process takes around 3 days to complete. I also did some post production using Adobe Audition and hosted the music on Soundcloud.
Poetry
The poetry was generated through a collaboration with GPT-3. The system is very simple and only requires at is most basic a simple prompt such as “A poem about RoboCop by Frank O’Hara.” The model has a series of fine tuning settings that can be adjusted to impact the output of the text also.
Posters
The posters are created in collaboration is a few of machine learning models, primarily Letent3Visions, and Halluzinator. The are very powerful text-to-image models using CLIP from Open AI and a model specific GAN. I provide a series of prompts and keywords to kick of the making and then the GAN works with/against CLIP to create and then identify images and compare them to my prompt for accuracy. Each time attempting to create something that the network thinks satisfies my prompt is some way.
