Aleatoric is an experiment in artistic intentionality, digital scarcity, and serendipity where I literally make art in my sleep: my sleep talk babble is auto-transcribed, auto-visualized, and minted into an NFT the morning-of. As the artist, I don't know when pieces will be minted, what they're titled, or even what they'll look or sound like. https://aleatoric.art In creating Aleatoric I wanted to tie the existence of a digital object—something with no inherent scarcity—to an unpredictable, naturally scarce event in the real world. By recording my sleep talking every night, the frequency and content of the work is entirely left to chance, taking the traditional definition of aleatoric music—in which some element of the composition is left to chance—to its logical extreme. Further interpretation of the sleep talking event is done automatically by machine, in which I have no real creative control. In many ways the specific transcription and visual interpretation of the AI models reflects back onto me, in effect telling me what I dreamed about that night. Much of dream interpretation is subjective, and the models provide a rather concrete interpretation of semantically vague mumblings. Learn more in our interview in the Zora Zine: https://zine.zora.co/aleatoric-nft-project Nightly, my sleep is recorded by the Sleep Cycle app, which uses its own models to segment and identify audio events throughout the night. If and when it detect me sleep talking, I share the sleep talking recording to my laptop and run a simple script that uses OpenAI's Whisper to transcribe the text, uses a few-shot gpt-4 model to generate a full stable diffusion prompt from the initial transcription, pipes that prompt into aleatoric-deforum (a slightly customized version of deforum), and mints the resulting asset as an NFT which is auctioned from $0.