AI Collab.
Freestyle H-ai-ku
“Freestyle H-ai-ku” is an AI-enhanced exploration of my Freestyle Haiku poetry. This project utilizes AI as a collective unconscious that not only intensifies the expressiveness of my poetry but also unveils hidden truths and newfound meanings within my words.
Candles Flicker (Token ID: #1)
Inspiration
“Freestyle H-ai-ku” is an AI interpretation of my ‘Freestyle Haiku’ poetry (freestylehaiku.com) that I’ve been writing since 2010. As a poet, I use haiku to crystallize my experience in words, using my interpretation of the free-verse haiku tradition, which was a more modern adaptation of the traditional haiku tradition. In Freestyle H-ai-ku, I use various AI collaborative tools to further evolve my concise, expressive poetry, and in the process, reveal hidden truths.
“What distinguishes a haiku is concision, perception and awareness… A haiku is a short poem recording the essence of a moment keenly perceived in which Nature is linked to human nature.” – Cor van den Heuvel, The Haiku Anthology, 1999.
In this project, I use AI as a sort of ‘collective unconscious’ to interpret the meaning of my poems, allowing me to see my own words with new meaning behind them. While my poems were the crystallization of my experience, their interpretations (and the interpretations of those interpretations) are abstracted through the lenses of modern culture: they are crystallized and interpreted through various AI models. The results are an amalgamation of the collectively perceived human experience, created by AI, based on the individually perceived human experience, my own. The resulting transformation is from text to image, from my individuality to a modern universality, and from my heart to the viewer’s mind.
“The art of poetry is beyond words… All writers look for a way to shape a group of words into an experience larger than its parts with images that will resonate long after they are read.” – Jane Reichhold, The Haiku Handbook, 1985.
The Collection
“Freestyle H-ai-ku” is a collection of AI interpretations of my haiku poetry, imagined as high-resolution AI paintings (7200 x 7200 pixels / 24” x 24” @ 300dpi).
NFT titles are taken from the poems, the NFT description is the poem and the Transformer LLM AI’s interpretation of the poem, and the traits are broken up into four sets: one being meta about the poems and the prompt style used (poem-number, date, LLM abstraction type, etc). The other three sets are keywords: the author’s tags from the poem, the keywords created during the LLM’s analysis of the poem, and artist-provided emergent traits based on the final image / end result of each collaboration.
Freestyle H-ai-ku #2 – #10
The first images in the series are quite varied, corresponding to the different themes and emotions in the poems that the artworks are created from. The rest of the series continues this trend, and while there may not be a visual cohesion to a random sampling of works, these are more than just images.
The story of their creation, their relation to one another in theme, time and emotion, as well as the poems themselves embedded in each of the artwork’s data all contribute to a cohesion of the project that takes effort and attention to appreciate.
The Process
It all starts with a poem, directly copied/pasted from my blog, freestylehaiku.com. I supply this to a transformer-based language model with specific prescribed instructions and expectations. The resulting text response is precisely tuned to the various inputs needed for a custom diffusion model. The resulting images from the diffusion model are then supplied to a GAN (generative adversarial network) designed for image restoration to increase image resolution and enhance detail. The GAN-generated images are assessed, and if they meet the artist’s approval, selected outputs are supplied to a CNN (convolutional neural network) to further enhance the image resolution, detail, contrast, and grain. If images do not meet the artist’s approval, the process cycles back to the LLM to generate an alternate, more abstract prompt. Once an image with promise is created, I pull it into an image editing suite to further edit and polish the work. All in all, “Freestyle H-ai-ku” uses 4 AI technology types: Transformer, Diffusion Model, GAN, and CNN, each individually trained on separate public data tailored to the AI’s usage.
Simplified workflow, also demonstrating the alternating perspectives during creation.
Individual vs Collective Experience
I find this project particularly interesting due to the interplay between the individual/subjective and collective/universal experience that influences the art at various points during its creation. The diagram above simplifies the process, but the full process that every piece in this series is created through is much more elaborate.
I approached this project not to create AI art, but to have AI interpret my poetry and transform it into something beyond my capabilities – something with a perspective larger than myself, literally as wide as the various AI’s training sets.
It is my hope that the art is viewed as a true collaboration between my poetry (frozen moments of myself as a poet), various AI technologies and tools, and myself as an artist.
The Smart Contract
I am a poet, I am an artist, and I am also a programmer. The smart contract that this collection uses is custom and unique, inspired by my desire to create and promote modular and composable blockchain assets. The contract design is such that it will allow on chain storage of many typically off chain data pieces. Due to the nature of this project as a collaboration (and that the diffusion model is about 8 gigs), most token data is instead stored on Arweave – a crypto-powered, decentralized storage network (technically a blockweave).
The poems are the most significant exception: they are individually stored in each token’s records, and they are composable to other smart contracts. I’ve always published this poetry for free, and their proper inclusion in this project was a crucial element to me. By using this contract, it allows a preservation of this poetry on chain in an easy to access and composable way. Token ID’s relate directly to the original poems, and from the contract, getDescription(_tokenId) will return the poem (and only the poem) as a formatted string, for free, forever.
Another fun aspect of the contract is a declared maximum collection size of 10^12 tokens. To me, this is important for two reasons: it’s a motivating element to try and write more poems and use AI’s to interpret and imagine them (I’m not fast, in 10+ years I’ve written ~1.3k poems). This massive maximum edition size is also a rebellion against the notions of predetermined scarcity that feel more commercial than artistic. Someday, I will no longer be able to write poems and use AIs to create new tokens, and when that happens, the collection size can be reduced (by myself or whoever manages my art after I’m gone). Until then, however, I plan to write as much as I can and create AI collaborative art based on it.
Prints Coming Soon!
Signed prints at 24 x 24 inches (the artwork’s natural size) will be available to order starting in December 2023.
Token ownership of the artwork will be required. Single prints per artwork vs editioned prints per artwork is TBD.