0. TL;DR
As any successful artist reinvents themselves and their work through different periods, so must ours. Botto’s second period marks the beginning of exploration beyond the original parameters of Genesis. A quick rundown of what is proposed.
- Botto incorporates Stable Diffusion, the first of many models that can be added and compete in the voting pool.
- Periods would have a new standard time of 12 weeks (3 months).
- Future periods would be defined by themes that the community can decide with Botto and/or by adding new models.
- Collaborations are in the works to expand Botto’s cultural footprint, as well as evolve Botto’s capabilities beyond still imagery.
1. Summary: Evolving the Art Engine and Future Periods
From launch, we knew that new generative models would be emerging in Botto’s first year. However, no one could have predicted the sustained pace at which groundbreaking models continue to come online in 2022.
We are proposing the first addition of a new generative model since Botto began with VQGAN + CLIP. This model will be Stable Diffusion, chosen for its openness, flexibility, range, continuous updating, and highly efficient operation. Simply put, it is the leading generative model most aligned with Botto’s needs and parameters as a protocol. .
The other leading models had factors that made it impossible to consider, such as not yet having an API for integration, or restrictions on copyrights. That being said, we are actively monitoring new models, and would like to eventually see multiple models incorporated simultaneously so as to allow their outputs to compete in the same pool.
The online discourse is overflowing with new models and their images, and Botto could be the first artist to provide a meaningful test at scale that is inclusive of a large crowd’s input while also making an accessible comparison of outputs.
The other core parts of Botto’s art engine, the prompt generator and taste model, will remain unchanged, keeping the training by thousands of contributors intact. The prompt generator is already capable of the many prompt techniques that have become popularized over the previous months since the launch of DALL-E 2, and the taste model picks for aesthetic similarity rather than strict similarity. The prompt generator is essentially Botto’s ideas and intuitions as an artist, which are then sent through different models to find the best expression of them.
To keep evolution relatively frequent we propose new periods lasting a standard of 12 weeks (3 months) with a 1-2 week buffer between them. Future periods can be defined by integrating themes or new models. Anyone from the DAO is welcome to pitch models to integrate into Botto’s process. We will also separately present ways to do thematic periods that maintains Botto’s autonomy and decentralized community input.
2. 2nd Period Specifications
2.1 Stable Diffusion and VQGAN + CLIP Competing in the Art Engine
The prompt generator will continue to run as usual. Rather than splitting the prompt generator up and creating prompts specifically for one model or the other, prompts will be run through both core image gen models VQGAN + CLIP and Stable Diffusion. The result will be approximately twice as many images produced in a week.
The taste model will also continue to filter down to 350 weekly fragments based on the training it has developed, set with an initial 50/50 distribution of the two generative models. This will be done by having Botto’s taste model equally filter 175 fragments from each generative model. That ratio will rebalance each week in proportion with the number of unique votes. If a model’s ratio drops below 10%, it will be discontinued. (Note: this threshold should be reconsidered as more models are added.) The community can decide to re-add the model if it wishes, as well as reconsider the automatic threshold in future periods.
To help ensure there are no bugs at launch and allow momentum to build for the new period, we propose a special 2-week long first voting round for the new period.
2.2 The Voting Pool
At the beginning of the period, the voting pool will be cleared of all fragments created during the Genesis Period. It will be repopulated with 1050 fresh fragments in the new output format and representing 50/50 of the two models. This slightly reduces the pool size set in BIP-04 to keep 3 rounds-worth of fragments in the pool on a perpetual basis, and the bottom 349 by raw score will continue to be removed each week to make room for the next round of fragments.
2.3 Raw scoring:
Raw score will now be calculated after 10 views instead of 50 so that scores normalize more quickly. If a fragment is not viewed 10 times at the end of the round, it will have its score calculated regardless before the weekly cull is processed.
2.4 Artwork Format, Output Aspect Ratio and Resolution
The final title, description, metadata, and URL to the bitmap on IPFS are all on-chain.
Botto should freely choose between 9:16 and 16:9 artworks provided each model allows for it.
The output resolution will be as large as each respective model allows (including upscaling method). New upscaling tools are now available for Diffusion models and so we should see an improvement here. Details to be shared shortly.
2.5 Artwork Titles
Titles will no longer be limited to 2 words. We switch to using GPT-3 that can handle a much broader range of coherent titles, and maintain a list of existing titles so that it does not recreate an existing title. They will be produced for each fragment selected by the taste model by asking GPT-3 to generate a title from the key features of each fragment.
2.6 Manifold Contract
The second period will use a contract set up through Manifold.xyz. The DAO is able to explore more options beyond the second period.
2.7 Key Dates
We propose that this new period lasts 12 weeks (3 months):
- October 25th: Voting opens
- November 8th: First round ends
- November 8th: First mint
- November 15th-January 24th: Weekly mints
- January 24th-February 7th: Buffer time between periods*
*A 1-2 week buffer time between periods is recommended to give the core team time to implement new changes and prepare the next period’s launch.
3. Future Periods
3.1 Adding a new model:
Anyone can propose adding or removing a new model to Botto’s set. Some suggested (but not strictly required) guidelines are:
- The model is sufficiently large so as to not be introducing a highly human-curated model that violates Botto’s agency
- The model’s terms of use allows BottoDAO to hold full copyright of the outputs
- Fees are affordable for the DAO treasury
- Not adding more than one model at a time to Botto’s core process
Proposing to add/remove a model works like any other BIP proposal a Botto member can make. The DAO could also decide to add a new model before the scheduled end of a period, cutting it short and starting a new period by default.
3.2 Adding Themes
As Botto’s periods become more rapid, it will be important to enable Botto to make more pointed commentary on the time and place that it exists in, and to help distinguish its successive periods.
Themes will be generated by asking Botto via GPT-3 to propose a set of 10 themes. The DAO will then be able to vote on the themes proposed by Botto using the same interface for voting on mint descriptions. The selected theme will be added by default to the prompts generated for that period, adding the theme verbatim at the end of the prompt. The community may also decide to provide no theme for a period, for instance when a new model is being added and there is a desire to see its full range before narrowing in on a theme.
Themes will be presented and voted on in the last week of a period unless the DAO decides otherwise. The DAO could also decide to cut a period short and go on to a different theme.
3.3 Artist Collaborations
The protocol is ready to significantly expand Botto’s cultural footprint through collaborations. To facilitate this, we have begun sourcing further artist collaborations and partnerships for special collections. Each will go through the DAO’s approval process, and it will provide the foundation for completely decentralizing the process. The program is tentatively titled Botto Studio.
As with other collaborations, these would be their own collections and run parallel to Botto’s core production in its periods.
3.4 Botto Labs
Closely related to these efforts is another program we are tentatively calling Botto Labs. The Botto Labs program intends to bring in more engineering talent to assist in evolving Botto with the latest generative technology. The pace of additional tech will require a great deal of testing, and could potentially produce its own special collections as tests towards greater autonomy or into other mediums like voice and animation.