The AI algorithm Botto has earned around $1.3 million from its first six works in the form of NFT. The auction is taking place on the marketplace SuperRare, reports New Atlas.
When creating a painting, Botto generates random strings of words and passes them into the neural network VQGAN, which, from examples of famous works of art, creates images.
After this, the algorithm CLIP determines how close the result is to the original string of words. If necessary, it makes VQGAN redo the image until it achieves a “passing score”.
CLIP also selects a two-word title for the painting.
To describe the image, Botto uses OpenAI’s GPT-3 language model. It generates up to ten different excerpts of abstract poetry, which the development team checks, edits and selects the most suitable option.
During a day the algorithm generates 300 images. Then the “taste model” selects 350 paintings each week and sends them to a vote by community members. The image with the highest score is converted into an NFT and put up for auction.
As of this writing, the project has completed six paintings. The first two sold for $300,000–$400,000, but the price of the most recent lot was $81,000.
The proceeds have been directed to the development of the project, which issues its own Botto token, used by community members to vote on what they consider the best options. After each auction, a portion of the tokens is bought back and burned.
In November, a 14-year-old artist in the United States earned more than $1m in 10 hours on non-fungible tokens.
In August, Fetch.ai launched an NFT marketplace for AI-created artworks.
Subscribe to ForkLog news on Telegram: ForkLog AI — all the news from the world of AI!