How Botto Makes Art
Botto’s creation mechanics are intertwined with novel technological and social paradigms, and there are various stages that Botto’s images pass through before becoming cemented in its canonical artistic legacy.
This article introduces the technical aspects of Botto’s art and describes how Botto’s work is curated by its community as a unique human-machine collaboration.
From Text to Image, Algorithm to Eye
Botto’s creation process begins with algorithmically generated text prompts, composed of random words and phrases, which are then fed into Botto's text-to-image models to produce visual outputs. This results in an enormous variety of potential images that can move through the process funnel from creation to shortlisting, and on to a single weekly canonical minted work.
Botto's art is consistently surprising, spanning endless artistic styles, subjects, and aesthetic strategies. As new members continue to join the DAO and vote, Botto’s work continues to evolve and change according to new tastes and artistic ideals among its voting community. Botto’s production cycle is demonstrated by the infographic below.
Botto’s Generation Models
Botto is trained on a combination of custom models created by Mario Klingemann and open source text-to-image models like VQGAN + CLIP and Stable Diffusion.
It is important that the text-to-image models be open-source so that they can be run autonomously with free commercial use, and that the publishing company cannot decide to censor Botto. Even more importantly, they are foundational models and do not represent any particular artist’s aesthetic, leaving room for Botto’s unique aesthetic voice to emerge. These models have been trained on more images than any single human could process in a lifetime, and are vast latent spaces representing the breadth of imagery contained on the internet. Klingemann designed Botto to search these latent spaces through its prompt generators and taste model initially using random words, phrases and features.
The art engine has evolved its aesthetic using the feedback of the community to find signals of “interestingness” within the latent space that could expand our notion of what is art.
As new open-source AI tools are made public, BottoDAO will continue to assess the viability of adopting new tools to continue the advance of Botto’s evolution.
Process in Detail
We can break down Botto’s process into two essential parts, the text engine and the taste model.
1. Text Engine: Because Botto needs text input in order to generate images, text is first generated by a custom text engine. In Botto’s earliest days, it was essentially a custom “random word generator” algorithm. Over time, community voting influenced these text prompts to become more refined.
If a particular type of text prompt led to images that received positive feedback, similar prompts (and the words and combinations within them) were more likely to be used again in the future. This feedback loop ensures that Botto's creations continue to evolve in alignment with the community's preferences.
2. Image Taste Model: Botto contains a taste model that filters its weekly production of 1050 images down to 350 works that will be presented to the community for voting. Similar to the recursive process within the text engine evolution, the votes on each week’s 350 images influence the outputs of the next week’s 1050 unfiltered images (which will then be filtered by the taste model, and then voted on, and then filtered, and then so on and so forth week-after-week).
As an example of this process, let’s say that this week one fragment is a fan-favorite and contains floral elements. It becomes highly voted on, and Botto will therefore include more floral words in the following text engine prompt generations and more floral imagery in the following image taste model filtering. More on this later.
The simplest way to conceptualize Botto is as an artificial intelligence artist that learns from human feedback and integrates that feedback into future outputs. It’s an autonomous, yet ever-evolving creative machine.
How Curation Works
Community curation is a critical part of Botto’s artistic process and growth, and the feedback Botto receives on its work from the community, via voting, serves as both guide and training for its future outputs.
Without feedback, Botto would likely produce work more or less at random. Voting provides direction for Botto to produce work that’s reflective of the artistic tastes of its community, forming what is essentially a collective creative practice.
Each week, a pool of 1050 new artworks (fragments) is created by Botto and enters a curation process where members vote on pieces they like best. At the end of the week, the top fifteen fragments go to a leaderboard, where the final piece is selected by the community and minted as an NFT.
After the final selection, 350 new fragments are added to the pool, and the 349 least popular fragments are removed (along with the piece minted in the previous week) to keep the total pool at 1050.
Voting Mechanics
The community votes for fragments in Botto’s voting app, either by accessing the available voting pools, which display two pieces side-by-side, or by directly voting on fragments inside the Botto Gallery.
In the voting pool, a user chooses which of the two pieces they prefer, and allocates a chosen number of Voting Points (VP) to that piece. Voters can also downvote a piece they particularly dislike, which will help train Botto to produce less work of that type.
To earn Voting Points, community members must either own select NFTs or stake the $BOTTO token.
For more information on Voting Points and $BOTTO, read:
Botto 101 - Part 5: The Economy and Staking
From Leaderboard to Auction
The weekly leaderboard features the top fifteen fragments by VP allocation and is open for voting for a period of twenty-four hours, typically starting on Mondays. After voting on the weekly leaderboard is conducted, the highest-voted fragment proceeds to auction on SuperRare, with the auction typically starting on Wednesday and ending on Friday.
At this time, the highest ranked fragment by VP from the leaderboard is what moves forward to auction, though there are ongoing discussions within the Botto community about potentially modifying this selection mechanism to incorporate the feedback of the crowd more vis-a-vis considering a fragment's Score as well. The Score reflects how often a particular fragment has been viewed and voted on by unique voters.
For more details on how the Score is calculated, see:
https://docs.botto.com/details/voting-mechanism#voting-mechanism
Botto Project's weekly fragment leaderboard may be viewed here: https://www.botto.com/leaderboard/fragments
Weekly Botto Curation Schedule
Tuesday at 16:00 ET / 22:00 CET:
Botto adds 350 new fragments and removes 349 old fragments from the voting pool, so a constant number of 1050 fragment always remains in the period’s voting pool
Tuesday-Monday:
Users vote for their favorite fragments in the voting pool or gallery.
Monday at 16:00 ET / 22:00 CET:
The top 15 fragments go to the leaderboard, and voters have 24 hours to boost those pieces with additional VP.
Tuesday at 16:00 ET / 22:00 CET:
The winning piece is selected, and the three-day auction is scheduled to begin the next day. Users have 12 hours after the leaderboard closes to vote on a description that will be minted as part of the piece.
Period Themes
Botto’s artworks are organized around 13-week periods, with each period based on a theme. These themes are chosen by voters from a separate pool generated by Botto.
The period themes add new artistic elements to fragments, and also give guidance for voters when choosing their favorite fragments. Each winning fragment has an accompanying description generated by Botto to describe its own work.
As of this writing, there have been 6 canonical Botto periods to date:
Period 1: Genesis Period
Period 2: Fragmentation
Period 3: Paradox
Period 4: Rebellion
Period 5: Absurdism
Period 6: Interstice
Further Reading & Resources
Botto 101 Continued
Get Connected
