Neural networks as a new way to create NFTs

Neural networks are one of the models for implementing the capabilities of machine learning and artificial intelligence. This is an old, large and rapidly developing market: in 2018 it was estimated at $10.1 billion, by 2025 the estimated volume of the sector could be $126 billion.

Until now, artificial intelligence and neural networks have been used mainly to solve corporate problems, such as optimizing and improving the performance of Yandex and Google search engines. In 2022, another trend has emerged, focused primarily on private users: AI (Artificial Intelligence) as a tool for entertainment and creativity.

A wide audience learned that neural networks can be used to create digital art objects, and this technology is closely intertwined with cryptocurrencies and NFTs.

Neural networks are a fairly complex technology that can be classified according to dozens of features. This article discusses programs designed to create objects of art - according to the basic principle of operation and interaction with the user, they can be divided into the following groups:

Neural networks of the “Text to Image” type work as follows:

This model is based on the principle of one of the most complex cognitive processes in the human mind. When a person reads a book, listens to a story, or carries out mental activity, associative visual images illustrating the incoming information appear in the brain.

Text to Image is the most common type of neural networks.

Another, although not as common, method of generating images is “Image to Image”. Sometimes it is also called "Drag and Drop". To create images in neural networks of this type, an existing image, sketch or photograph is taken as the basis.

For example, Artbreeder uses a combined method - an image supplemented with a text description. To generate an image, the service prompts the user to create a collage from the elements of the image base and a text description of what needs to be obtained.

There is also an experimental project Nerf in The Wild: the robot is able to create three-dimensional videos from several images of an object taken from different angles. You can’t use it for your own purposes yet, but this is one of the clear examples of the Image to Image model - or, in this case, Image to Video.

Chatbots are an alternative model of interaction between a person and a neural network to web services. The first primitive bots appeared back in 1966, but since then IT has made a huge leap forward: modern trained machines support conscious and intelligent conversation no worse than a real person. A new text chatbot from the developer of Open AI, launched in 2022, ChatGPT, is able not only to conduct a meaningful dialogue, but also to give a detailed answer to questions - even very complex ones. We found a small experiment that showed that sometimes this bot can be more useful than Google and can write small analytical articles.

Such chatbots are used not only for communication - over time they were taught to generate images in response to a user command. Open AI has such a project.

To date, many neural networks have been created for generating images – a lot of them appeared only in 2022, when the trend for AI and machine learning reached another peak. The Open Ai mentioned above, which offers similar features, is called DALL-E 2 and creates pictures based on a text description.

It is enough to enter a few words, and the artificial intelligence will offer four relevant photos compiled by it:

Or pictures, if the request is a bit specified:

Another platform called Deep Dream Generator is an example of a service that can create new pictures based on an uploaded image and a text description. This is what Van Gogh’s “Starry Night” looks like, passed through artificial intelligence: 

Another tool which can help to create pictures in a couple of clicks is MindJourney. It is also new: open beta testing began on July 13, 2022, as the developers announced on their Twitter. Testing is ongoing at the time of writing this paper and users are free to join.

Unlike previous services, MindJourney only works as a chatbot on the Discord platform - we will explain how to use it later.

Here is the official site of the project, where you need to register to create images: link.

The MindJourney team has prepared a small manual for the user - we have selected the most important rules from it:

So far, the service works only in English, which is also used to enter commands and descriptions.

Use the invite link to the official MindJourney channel to join the chat. Before that, you need to create an account in Discord - an email address is enough for it.

The project channel looks like this:

It has several chats, but to create images, we need a section called “Newby” - there are several of them, and it does not matter which one to choose. To start the bot, you need only one command:

A promt area will appear on the command line, where you need to enter a description of the future picture – to do it you can use a translator. The more detailed the description, the more accurate the picture will be. For the sake of experiment, we decided to give the bot our own task: to generate a picturesque forest landscape with a deer standing in the middle. A few details were added to the description:

After sending the request, it will take the bot about 60 seconds to generate four images. In the process, we were able to observe how each of the pictures was transformed, and the result turned out like this:

Now this picture can be used on any NFT marketplace, such as OpenSea or Binance NFT. On Binance, to be eligible to create tokens, you need to recruit at least two subscribers - you can invite friends or relatives, and then follow these steps:

Collectible tokens remained the most popular and sought-after niche in the NFT sector, including the last quarter of 2022:

This trend is likely to continue into 2023. But not only pictures or avatars can be collected: thanks to the skills of artificial intelligence and neural networks, users can create NFTs based on music, 3D models, and even poetry.

Open AI, which developed chatbots and services for generating images, presented another project at the end of 2022 - a neural network for creating 3D Point-E models. It works on the principle of “Text to Image”: a brief description must be entered in the field, on its basis the model will be generated.

This concept is very young and very difficult to implement - so the work of Point-E is still quite primitive:

The experience of previous projects of this type suggests that you just need to wait a bit, and over time the program will learn to produce full-fledged 3D models.

Music has long been sold in the form of NFTs - even Snoop Dogg has his own collection on OpenSea. To do the same, it is not necessary to create your own tracks from scratch: neural networks are successfully used for this.

To create music in the BOOMY neural network, you need to select only two parameters - style and theme. After generating the recording, the service offers to edit the composition and add vocals to it.

Neural networks can not only communicate, but also create entire works - essays and even poems. We found a service called Poem Portraits by Google that generates text from a base word and creates a card with your selfie.

AI Poem Generator works without a selfie - instead, it needs to set a few words and associations for a future poem.

However, collections and earnings are not the only area for which you can create NFTs using neural networks. Looka is a service that generates logos based on given parameters. A logo designed in this way can be digitized and thus protect the ownership of the image.

Artificial intelligence for digital art creation as a new trend in the NFT sector was predicted at the beginning of 2022. Expectations were justified - by the middle of 2022, user interest in image generators had increased like never before:

The most popular queries turned out to be the names of project companies offering their neural networks:

AI has become an affordable and very easy way to create NFTs. After the end of the bearish trend in the market and the return of the former demand for non-fungible tokens, art created with the help of neural networks can grow many times in terms of value - and this is definitely worth taking advantage of.