Experiments with neural networks: 5 recent cases

Contents:

Neural networks are actively changing the design landscape, and the discussion of their impact on the design profession has become relevant. We spoke with industry professionals who have already integrated artificial intelligence into their projects. They shared their opinions on the advantages and disadvantages of this tool, as well as its prospects for the near future. Neural networks can accelerate the development process and offer new creative solutions, but questions remain about the quality and originality of the resulting content. It's important to understand how artificial intelligence can complement, rather than completely replace, human design.

This article provides useful information on this topic. We'll discuss key aspects and provide valuable tips to help you better understand the subject. Read on to gain knowledge that may be useful in your practice.

• Texts for a German website from ChatGPT
• Experiments with fonts on Midjorney
• AI illustrations on Lamoda
• Sketches of tattoos from Midjorney
• Paintings in Stable Diffusion

Designer and co-founder of VOSK.

A few months ago, we began collaborating with a large German chemical distributor. The client tasked us with modernizing their outdated website and creating a new product catalog. They plan to update their identity in the future, so they requested that we define a global visual style through the updated website. This approach may not be the most effective, but it allows us to lay the foundation for further changes.

The main problem with the outdated website was that it was not adapted to modern requirements. Its appearance consisted of several pages of text and contained a very limited amount of useful information. This negatively impacted the user experience and reduced visitor engagement, which in turn affected SEO. To improve the site's effectiveness, it was necessary to update its design, improve the content structure, and provide users with valuable information that meets modern standards.

Misha Derichenko, responsible for user experience, decided to use ChatGPT for text processing. As a result, we received a concise and informative version, which we adapted for the website. This allowed us to significantly save resources by avoiding the need to engage a native speaker with experience in UX/UI copywriting. Thanks to this, the necessary information to get started was collected in just a few hours.

ChatGPT demonstrated high accuracy in wording and describing complex technical processes, such as chemical distribution. All generated texts were approved by the client's employees, who expressed no comments or questions. This confirms the reliability and accuracy of the information provided by the system, which is especially important in an industry that requires strict adherence to quality and safety standards.

The new visual style will form the basis for the upcoming rebranding, but we did not have sufficient time to test all the hypotheses. A neural network assisted us in this process, with which we created realistic product photographs and abstract 3D illustrations. The specific style will be determined during the development of the site, which will allow us to more precisely adapt the visual elements to the needs of our audience.

Product designer and design lead at Novado, specializing in digital environment research. We create innovative solutions that improve user experience and drive business growth. Our team focuses on analyzing user needs and applying modern design approaches to optimize interfaces and increase the effectiveness of digital products.

I became interested in neural networks when I saw images created using Midjourney. A friend of mine, a CG artist, sent me pictures of robots from Evangelion on Telegram, stylized for an Orthodox context. In response, I created my own versions, imagining them as powerful robots. This interesting play with visual imagery inspired me to further explore the potential of neural networks in art and design.

The result was abstract, yet at the same time extremely artistic. The ability to combine different values ​​and quickly form a concept for any idea amazed and inspired me.

At the time, I was actively studying fonts and typography and decided to experiment with generating letters. However, the results were disappointing: Midjourney couldn't handle the task, producing low-quality letter images. Cyrillic proved particularly problematic, as the neural network couldn't recognize it and, as a result, generated distortions instead of normal characters.

The unpredictability of the results creates space for creativity. I created many images for the search query "Cyrillic letters." Then I explored abstract letterforms, trying to find those that inspire me. I redrew them in vector format to improve my skills, understand the plasticity of forms and the logic of their construction. These practices help deepen my understanding of typography and develop a personal style in design.

The most amazing thing happened when I began combining letters into words. Abstract symbols harmoniously intertwined with metaphors that arose from their forms. This process transformed each letter into a carrier of deep meaning, creating unique images and associations. Words born from these combinations became not only expressive, but also saturated with emotional and intellectual content.

Neural networks create content that looks very familiar and similar to human works, but at the same time creates a feeling of unnaturalness and alienation. This paradoxical combination evokes surprise and attracts attention, while simultaneously alarming users. The development of artificial intelligence technologies poses new questions about the boundaries of the creative process and the role of humans in it.

My experience is a creative exploration aimed at finding ideas and working with references. While this approach may not seem the most traditional, it demonstrates the potential of combining a designer's professional experience with the capabilities of neural networks. Such combinations can become the basis for a breakthrough in the field of visual communication, graphic and typographic design, opening new horizons for creativity and innovation.

The Head of Design at Lamoda is responsible for design development and brand communications management. He plays a key role in shaping the company's visual identity, ensuring that design aligns with strategic goals and customer needs. His responsibilities include creating attractive and functional solutions that strengthen Lamoda's image in the marketplace.

Since March 6th, photographs generated using a neural network have been posted on Lamoda's page as part of an experiment. We aimed to reflect current fashion trends, combine them, and ultimately present something unique and innovative. Our goal is to demonstrate how modern technology can influence fashion and inspire users to create their own style.

Our design team carefully analyzed the latest fashion trends, characteristics of famous brands, and cultural influences, resulting in the creation of over 50 unique AI illustrations for the homepage. As part of the project, we also engaged major brands such as Ecco, Love Republic, Urban Tiger, and Rinascimento, adding additional value and variety to our work. These illustrations not only reflect current fashion trends but also highlight the individuality of each brand, making our project even more appealing and contemporary.

With the rise of artificial intelligence in the fashion industry, we began collaborating with Baon, using neural networks to create unique visuals. This was an internal project, and initially, we needed to master neural networks to ensure a high level of image quality. Our goal was to leverage the power of artificial intelligence, which enabled us to create impressive and original results.

The results were inspiring, so we decided to create AI-powered images for the homepage and showcased them throughout the week. On our "Lamoda in Touch" Telegram channel, we published content from our virtual editor, L Æ MØDA-11 (or Ela), who also generated content using artificial intelligence. This allowed us not only to diversify our visuals but also to offer our subscribers a unique experience interacting with modern fashion technologies.

In terms of conversion, many banners showed significantly better results compared to standard options. This case sparked active discussions in the design and IT communities. About 90 hours were spent on developing the images. Graphic optimization and a creative approach to design significantly increased the effectiveness of advertising campaigns, emphasizing the importance of a professional approach to creating visual content.

Image generation provides a unique opportunity to not worry about time and resources. This process is both exciting and challenging. In the future, in my opinion, many designers will increasingly delve into creating content using artificial intelligence technologies. AI will become an essential tool for design professionals, opening up new horizons and opportunities for creativity.

Graphic designer and student of the Krasnodar Institute of Contemporary Art (KISI). I specialize in creating visual content that effectively conveys ideas and emotions. My education at KISI allows me to deeply study modern trends in design and develop skills in various fields of graphic art.

In December, I monitored the Profi platform in search of interesting orders. Among them, I came across a request related to tattoo sketches created using neural networks. The client was looking for original images of famous rock stars as animals.

I researched portraits of the Motörhead frontman and used the query "monkey-like" in Midjourney to generate images. I experimented with different wording to understand how the neural network responded to different synonyms. The neural network responded better to some synonyms than others, depending on the dataset used and how it was processed. The key was choosing the right reference that would allow the neural network to accurately capture the character's features and transfer them to the generated images. This allows us to achieve higher quality and more realistic results in content generation.

The client approved the result and received a sketch, and I gained valuable experience.

After a month of working with a colleague, we began creating illustrations for a board card game. To simplify the search for compositions and diverse character styles, we used image generation. Although the results could have been used immediately, our perfectionism prompted us to redraw the generated images and create additional variations. While the final version of the project did not include any images generated with Midjourney, we used compositional references based on these generated works.

I often experiment by combining posters from different eras and cultures in search of new art forms. The results are always unexpected. When I try to combine themes of astral projection, Soviet constructivism, and elements of Nintendo games, the final piece, created in just a minute, is truly impressive. Such experiments open new horizons in visual art and allow us to see familiar things from a different perspective.

I'm drawn to Midjourney because it offers the opportunity to create unique images without having to delve into the complexities of artificial intelligence. This service makes it easy and convenient to generate visual content based on your own prompts, making the process creative and accessible. Users can focus on their ideas and inspiration, not on technical details. Midjourney opens new horizons for creativity and self-expression, facilitating interaction with AI in the field of image generation.

Following the scandal with realistic images of the Pope, the Midjourney platform limited free access. This inspired me to move on to studying another neural network—Stable Diffusion. This neural network offers significantly more control over the image generation process thanks to its open source nature. I can now customize the results more flexibly and experiment with various parameters, making the process of creating unique images more engaging and efficient.

Stable Diffusion stands out for its powerful training capabilities. Unlike Midjourney, this platform allows users to create their own models. For example, you can upload around 30 photos of your face and use them to generate images of you in various poses and situations. This allows for unique and personalized results, making Stable Diffusion particularly appealing to those seeking flexibility and a personalized approach to image generation.

A similar approach can be applied to other visual fields. For example, I studied the work of Japanese poster artists such as Yusaki Kamekura, Kazumasa Nagai, Shigeo Fukuda, Koichi Sato, and Ikko Tanaka. These Japanese designers are captivating with their masterful use of color and abstract forms. I'm currently exploring adapting their poster styles to the Stable Diffusion algorithm and combining them. I learned about the methods for implementing these ideas from training videos on YouTube by Khachatur Avanesyan.

Sharing your experiences and mistakes during the learning process is extremely important. My joint study with colleagues and friends significantly accelerated my progress. This strategy also contributed to the rapid development of neural networks. If you are interested in this topic, you can contact me on Telegram to discuss it.

Illustrator and teacher of animation and illustration at the HSE School of Design and BBE. I specialize in creating visual content and teaching students the basics of illustration and animation. My experience and knowledge help develop creative skills and an understanding of modern design trends.

Modern generations of artificial intelligence actively demonstrate the aesthetics of "this was created by a neural network", which makes them conceptually less interesting. As final visualizations, such works don't meet my expectations. However, the use of AI in the creative process brings a necessary element of surprise and variety to composition, subject matter, and color schemes. This creates unique opportunities for artists and designers, allowing them to experiment and find new approaches in their work.

In my first experiment, I used Stable Diffusion to generate scenes based on paintings. This series is dedicated to the future and covers various themes, such as prayer and hope. I formulate prompts for Stable Diffusion based on the figurative representation of the theme. For example, for a story about prayer, I used the query "girl praying in a blood bath," and for a story about hope, I used "tender lovers in pink tones."

I prefer working with oil on canvas rather than wasting time creating color sketches and searching for nuances. So, I created multiple versions for each project, combining elements in collages, making compositional adjustments, and then transferring the results to canvas. This approach allows me to implement ideas faster and focus on my main work.

When working with expressive materials like ink on paper, it is necessary to create many iterations to achieve the desired result. For example, it might take about 150 attempts to create a unique spot before finding the perfect one. My attitude toward artificial intelligence is similar: 99% of generated content may be inappropriate, but the remaining 1% can be a real find. It is this valuable material that is used in further work.

I've been experimenting with digital illustration using an approach where AI creates a narrative and then refines it. For example, I worked with the theme of "female friendship" and used the prompt "two girls in a pool holding hands." I loved how the AI ​​created interesting lighting, which I was able to adapt and improve in my work. This method allows me to combine the creativity of AI with my artistic vision, which makes the creation process more fun and inspiring.

I recently experienced the generation animation and was interested in how artificial intelligence interprets morphing, turning one image into another. However, the animation lacks finality; I would like to refine it using rotoscoping in my own style, but I haven't found the time yet. Nevertheless, as a tool for generating ideas in the field of morphing, AI-generated animation shows excellent results.

Learn more about design by subscribing to our Telegram channel. We share relevant news, tips, and inspiration in the world of design. Don't miss interesting materials and useful resources. Join us!

Articles on Neural Networks in Design

Neural networks are becoming an increasingly important tool in the design field. They can automate routine processes, analyze vast amounts of data, and propose innovative solutions. In this category, you'll find articles that explore the application of neural networks in various aspects of design, including graphic design, web design, and UX/UI. We'll discuss how neural networks can improve creative processes, increase designer efficiency, and offer new ideas. Learn how artificial intelligence technologies are changing approaches to design and opening up new horizons for creative professionals.

• Ask the expert: can a neural network replace a human designer?
• Test: neural network vs. designer
• Media artist Vadim Epshteyn on how he uses neural networks in his work
• Artists on neural networks as a trend in illustration

A court has ruled to ban Meta Platforms Inc. from operating in Russia, including the Facebook and Instagram social networks. This decision is based on charges of extremist activity. The ban impacts user access to these platforms and highlights the importance of complying with digital and social media laws.