Modern forms of generative art
Generative art meets crypto art
Last updated
Generative art meets crypto art
Last updated
In the generative art process there are always two roles:
the composer, who designs and implements the code that generates the artworks, and
the curator, who chooses the most interesting outputs from the many generated ones.
We will see how these roles, and in particular the curator’s, can be played by different actors in different forms of generative art.
These first digital pioneers - like the 3N of computer graphics Georg Nees, Michael Noll, and Frieder Nake - used archaic programming languages like Algol and Fortran. The corresponding code generated punched tapes. When read by a computer these tapes printed an output using some form of drawing machine, like the Zuse Graphomat Z64 used by Georg Nees and Frieder Nake.
Given the inherent limitations of this early creating process, the output was typically a single piece.
In this form of generative art, the composer is the artist (the programmer), while the curator is in fact the machine, since it was responsible for picking a single output from the many possible ones (assuming randomness was used in the code, which was frequently the case).
In modern day generative art, due to the abundance of computational resources (computing power and memory), a generative artist typically runs the code and generates a large, potentially unbounded number of outputs. Then, the artist picks the best highlights, often a single output, and possibly mint it as an NFT on a crypto art marketplace.
In this generative art form, the artist is both the composer and the curator of the generative art process.
Recently, generative art NFT marketplaces like Art Blocks on Ethereum and fxhash on Tezos allow for something different. The artist-programmer does not simply create an artwork, but writes a program, which is stored on-chain and generates a very large number of different outputs.
The algorithm uses a hash (a hexadecimal string encoding a block of information) as a source of randomness and the output is totally determined by the hash. Therefore, different hashes will produce different outputs. The collector chooses one of the available projects, for which there is at least one example of output.
The collector buys an execution of the chosen code at a given price.
The code generates the artwork and the associated NFT is transferred to the collector’s wallet. Neither the artist nor the collector know in advance what the output will be.
Basically, the original idea of this generative art form is to distinguish between the score (the code) and the execution of the score (the output of the code). Generative art thus becomes a performance, just as when we attend a concert knowing the score but ignoring what the musicians’ performance will be like.
As Tyler Hobbs notices:
This form of generative art introduces the new demands of consistent quality and high variety, while maintaining the existing need for unity across all output from a program.
Quality and variety are somewhat opposing forces and it is hard to maintain both for a large number of executions. For this reason, the artist must limit the search space of the outputs of the code, trying to maintain a good average quality among the outputs and avoid outliers.
Similarly to the making of short-form generative art, also in long-form generative art the artist-programmer plays both the roles of composer and curator of the generative art process. However, the curation step is fundamentally different.
While in short-form generative art the artist cherry-picks the best outputs to highlight them, in long-form generative art the artist needs to be very skillful to inject the curation step into the code. In some sense, the code must have the knowledge of where the good outputs are in the search space.
There is, however, a hidden pitfall in long-form generative art. In order to encode the curation step inside the code, the algorithm needs to artificially bound the output search space. While this limits the possibility of getting a low-quality result, it also cuts off the chances of generating unexpected high-quality outputs.
Notice that so far the role of the collector was not an active one. In particular, the collector takes no part in the creation process. In single-form and short-form generative art, the collector passively buys what the artist offers. In long-form generative art, the collector mints the artwork, but the actual artwork is chosen by the algorithm.
In unbounded-form generative art the algorithm is free to explore the entire search space of outputs. The collector can generate an unlimited number of outputs and pick one among them for it minting as an NFT.
In unbounded-form generative art, the composer is the artist and the curator is the collector.
QQL is an unbounded-form generative art project by Tyler Hobbs - a visual artist who works primarily with algorithms, plotters, and paint - and Dandelion Wist Mané - a dancer, engineer, entrepreneur, and generative artist. The two artists describe the project as follows:
This collaboration is intended to provide a new way to mint NFTs that celebrates emergence, unpredictability, and happenstance over forced rarity. We want to encourage collectors to explore the edges of the algorithm, play a role in the output, and take agency to become a co-creator. We want collectors to view their engagement as an adventure and make a creative contribution to the art. Adding a curation step by the collector also allows the generative algorithm to take more risks and explore a more interesting potential output space. We trust collectors to seek out and identify the truly special outputs that emerge. With this approach, the collector is now the curator.
The last form of generative art, dubbed community-form generative art, leverages collective intelligence.
In this case, the curator is a community of individuals and organize themselves, possibly in the form of a Decentralized Autonomous Organization (DAO), to guide the generative process and choose the future outcomes.
A popular example is Botto, a decentralized autonomous artist. It creates works of art based on collective feedback from the community. It is an AI that creates works of digital art based on collective feedback from its vast community. In some sense, the human participation is what completes the artificial Botto as an artist.
More specifically, Botto's art engine has been trained on millions of images. It uses a combination of AI algorithms, including Stable Diffusion for image generation and GPT for text generation, plus a number of custom augmentations. Botto creates new unique images every week, all untouched by human hands. It controls itself generating its own prompts, doing its own filtering of images, and writing artwork descriptions. Though, Botto is still dependent on humans for many things, such as getting feedback to tune its aesthetics, adding new generative models, producing exhibitions and collaborating with other human artists.
Each week, Botto presents a number of promising pieces for consideration by the Botto community. The presented artworks are not yet considered final works, in fact they are called fragments as they are still unproven. The community tells Botto what it considers good art, and Botto perpetually evolves from the feedback. The most popular fragment is minted each week as a final artwork on SuperRare and sold at auction. A share of the total revenue is held in treasury controlled by the community and the remaining is distributed each week among contributors proportionally to voting points spent out.
