How to Use Durer.ai (Exactly.ai)

Contents:

• How to use ready-made models in Exactly.ai
• How to create your own models in Exactly.ai
• Terms of Use
• Competitors
• Neural Network Art at a London Exhibition
• Exactly for Experimentation and Research
• Application in Jewelry

Exactly.ai is a powerful neural network for generating images that has attracted users thanks to its intuitive process for training its own models. Users simply upload their work, and the artificial intelligence creates unique images similar to those submitted. The neural network was previously known as Durer.ai, and you may already be familiar with its capabilities. Using Exactly.ai opens new horizons for artists and designers, allowing them to experiment with creative ideas and gain inspiration from original paintings created by artificial intelligence.

We studied the principles of neural networks and conducted interviews with designers who use Exactly.ai in their work.

How to use ready-made Exactly.ai models

The service's basic functionality is available free of charge; all you need to do is register on the Exactly.ai website. After completing the registration, a link to access the workspace will be sent to your email address.

In Exactly.ai, as in other popular neural networks for generating images, users can create images using Ready-made models. To view the list of available models, simply open the menu in the upper left corner, located above the search field. This menu features dozens of different models designed for creating hand-drawn illustrations, photorealistic images, and imitations of classical art. Each model offers unique capabilities, allowing users to realize their creative ideas and achieve high-quality visual results.

Let's look at a model that creates watercolor illustrations. As an example, use the search query "brown pony walking on the beach." The result is stunning and conveys the atmosphere of tranquility and natural beauty typical of watercolor paintings. This approach allows us to create a unique image that combines elements of painting and digital art.

The neural network is capable of processing queries in Russian, but the recognition level often leaves much to be desired. For example, when asked for "brown pony walking on the beach," the model returned a result that was not quite expected. This highlights the need for further refinement and improvement of algorithms to improve the quality of processing Russian-language queries.

The principles of query generation in Exactly.ai are similar to those used in other neural networks. It is important to consider several key aspects. First, you should formulate your request clearly so the system can correctly interpret your intent. Second, using specific and detailed phrases will improve the accuracy of the response. In addition, it is helpful to ask questions that encourage detailed answers rather than just monosyllabic ones. Finally, experimenting with different wording will help you find the optimal approach for obtaining the most relevant results. These principles will help you get the most out of Exactly.ai.

• Specify the main object in the future image.
• Add a description of it or the action it is performing.
• If necessary, add more detail: describe the location of the action, secondary objects.
• Avoid the particle "not" and other negations. Negative prompts can be specified in a separate field: to access it, click the icon with three dots under the field for entering a regular prompt.

There is no need to specify the style in the request, as it is already defined. We tested the neural network with the queries "brown pony walking on the beach 3D" and "brown pony walking on the beach oil," but the neural network was unable to generate 3D images or oil paintings within a model focused on watercolor illustrations.

Image generation is possible based on your own sketches using the tools available in the right-hand menu. In the interface, you can create images, adjust line thickness, and delete unnecessary elements. There's also an option to upload existing sketches, significantly simplifying the image creation process.

We created a sketch and selected a model for drawing pencil illustrations. The topic for the work was prompted "ginger cat, sleeping."

Another option is to edit a specific area of ​​the image. Below the search field, there is an icon resembling the Lasso tool. By clicking on it, you can select the desired area of ​​the image, start re-generation, and the neural network will update the selected fragment. This feature allows you to fine-tune the image, improving its quality and meeting your expectations.

Neural networks have an interesting feature: when opening a new model, they automatically suggest a prompt based on their algorithms. This feature can be useful if you need fresh ideas for inspiration. For example, when using a model specializing in medieval drawings, the service suggested the query "a man pierced with swords." The result was quite impressive.

How to create your own models in Exactly.ai

To open the models menu, click on the empty field located at the top of the interface, labeled «My model». After this, by clicking on the plus sign, the user will be able to upload their images. This process allows you to quickly and easily add your own images to the system.

Neural networks require 5 to 15 images to operate. As an example, we have uploaded ten posters created by Toulouse-Lautrec.

Creating a model takes only a few minutes. Once the process is complete, you'll receive an email notification. You can work with your model just like any other: create images based on a text request or using a sketch. For example, for the query "person sitting at a table with a cup of coffee" we received the following results.

The Exactly.ai website states that when creating a model based on their own images, the copyright remains with the owner of the original images. This means that the generated illustrations can be used at your discretion, including commercial applications. Thus, users can freely dispose of their works, which opens up new opportunities for creativity and business.

Read also:

Neural networks and copyright: lawsuits, restrictions and uncertainties

In recent years, neural networks have become an integral part of many creative processes, but their use raises many issues related to copyright. Legal cases concerning intellectual property and neural network technologies are becoming increasingly important. There are significant legal limitations that must be considered when using such technologies.

One of the key issues is who is the author of a work created by a neural network. This creates a legal gray area where it is difficult to determine who owns the rights to the result of artificial intelligence. Litigation on these issues is becoming increasingly common as copyright holders attempt to protect their interests.

It is important to understand that the use of neural networks in creative work requires a careful approach and an understanding of the possible legal consequences. Due to the development of technology and changing legislation, it is necessary to monitor current changes in the field of copyright and neural networks to avoid legal problems in the future.

Terms of Use

In the free version, users can create up to ten custom models and generate up to 50 images daily. Access to the full library of Exactly.ai models allows you to draw, upload your own sketches, and edit individual image fragments. All works and models are saved in the history, which makes project management convenient. Generated images are 1080x1080 pixels in size and include the service logo on the side.

The paid version allows you to create an unlimited number of models, generate any number of images, and obtain high-resolution images without a logo. The subscription costs $50 per month.

Competitors

Many neural networks are focused on the use of ready-made models, among which are such platforms as Kandinsky and Starryai. In addition, there is a service called Recraft, which focuses on creating hand-drawn images. However, the ability to create your own models in these neural networks is not available. In contrast, Midjourney and Stable Diffusion offer this option, although the process of creating custom models is more technically challenging than with Exactly.ai.

Exactly.ai offers the unique ability to quickly and easily create models directly in the browser with just a few clicks. This allows users to achieve high-quality results suitable for professional use. The simplicity and accessibility of the process make Exactly.ai an ideal tool for developing AI models.

Neural network art at the London exhibition

An artist, designer, and art director are key figures in the field of visual art and design. An artist creates unique works that reflect their creative vision and emotional state. A designer, in turn, focuses on functionality and aesthetics, developing visual concepts for various projects. An art director combines these two disciplines, guiding the creation and implementation of artistic ideas, ensuring a harmonious blend of art and design. The professions of artist, designer, and art director require high qualifications, creativity, and the ability to work in a team, making them indispensable in the modern world of art and design.

I noticed a project developed by Russian specialists that offers neural network training based on their own work. Unlike Midjourney, where the training process is more complex, this project offers an intuitive interface, significantly simplifying interaction. The results of using the platform look promising.

I used Midjourney to generate images for my projects, mainly for experimental purposes. We recently generated preview images for case studies for event organizers who lacked high-quality photography and were striving for a unified visual style for their showreels. This process allowed me to not only explore the possibilities of image generation but also introduce creative elements into the visual design.

I've been familiar with Exactly.ai for a long time—I used it when it was still called Durer.ai. I was interested in testing the training of models based on my typographic assets. Neural networks are known to sometimes struggle with letters and words, especially when it comes to meaning. Although Durer didn't always perform well, it was able to generate interesting variations of other prompts that were more consistent with easily understood images of people and animals. As a result, I created a series of graphic portraits of famous rap and jazz artists. Using Exactly.ai in this process demonstrated how even complex problems can be solved creatively using artificial intelligence.

I trained several models, and one of them became the official Exactly.ai model, available to all users - Black and white scanned plants. This model was created based on graphic prints that my wife and I prepared for an exhibition. We used various natural materials from the area where we were and scanned them in Threshold mode. This allowed us to obtain black-and-white images without gradations. The model learned to generate similar "scans," and we were pleased with the results. Generations from this model were presented at an exhibition in London, organized by the Exactly.ai team.

Exactly for experiments and searches

Designer and illustrator, as well as a lecturer at the School of Design at the Higher School of Economics. He has significant experience in graphic design and illustration, teaching students modern design methods and techniques. He participates in various projects aimed at developing creative skills and visual thinking.

I haven't worked with other neural networks, but I know that Stable Diffusion has similar capabilities. However, this requires a significant level of knowledge. The process involves uploading your own work, after which the neural network creates a model based on your unique style. Often, a short prompt of a few words is sufficient. You can also formulate more detailed requests, for example, specifying specific materials.

If you have five images in your style, Exactly often successfully continues the sequence based on the suggested prompts. This feature is most effective when the requested works are in the same genre. For example, if you provided images of animals, it can easily create new images of other animals. If you uploaded landscapes, you can request the creation of new locations. I focused on the experimental search for new forms. This approach is less applicable to illustration, since the results can be unpredictable. Nevertheless, such work allows me to express myself as an artist who blends various cultural and figurative phenomena, creating unique works of art.

I implemented a large-scale project for GES-2 called "In Full Voice." This complex and multifaceted project involved volunteers rehearsing popular songs in a karaoke format over the summer. These melodies evoke nostalgia and sometimes make one feel embarrassed when admitting their love for them. As part of the project, I created a video sequence to accompany more than 20 songs, each of which required a unique approach. Some compositions evoked a desire to play with form, others - to create an atmosphere for a shared experience of emotions. The result was a large concert in the fall, which became one of the most unusual and memorable experiences of my career.

I've tried creating custom images, but they haven't been successful. While Exactly is indeed capable of handling uploaded images more accurately, complex tasks with multiple inputs often make it difficult to achieve the desired result.

Application in jewelry

The creator of the Besotted Jewelry brand actively runs a Telegram channel dedicated to experimenting with neural networks. On this channel, he shares his discoveries and ideas that connect the world of jewelry art and modern technology. Using neural networks in jewelry design allows you to expand the horizons of creativity and offer unique solutions for jewelry connoisseurs. Channel subscribers can follow the process of creating new collections and get inspiration for their own projects.

I generate jewelry sketches using neural networks. I've previously worked extensively with Midjourney, but I also had the opportunity to test early versions of models such as Stable Diffusion, DALL-E 2, and Kandinsky. I've now switched to Exactly.ai. This solution not only allows me to create unique designs but also trains models based on photographs of my work, which sets Exactly apart from other neural networks. I've now accumulated several thousand generated sketches, and I've decided to take a break to analyze the results and identify the most interesting and promising options.

The sketches presented demonstrate a wide variety of ideas and stone setting options. In most cases, the neural network correctly placed the prongs, but there is significant variation in style. I would like to achieve greater unity in the design.

Midjourney creates high-quality and realistic images, but my purposes don't require detailed rendering. The main thing is to see the idea, image, and form in the sketch. That's why I decided to use Exactly.ai. This tool is especially useful for those with minimal experience and an understanding of what's possible. Some images look impressive, but I recognize that implementing them can be challenging or even impossible. However, there are plenty of resources that can be used effectively.

I recently trained a model in Exactly using generated images from Midjourney. The results were impressive: by loading four variations of a single ring, I received sketches of earrings and pendants in the same style. This process demonstrates how modern technology can be used to create unique jewelry by combining artificial intelligence and creativity.

I refine the images in Photoshop, after which I create 3D models. Jewelry based on designs from Exactly.ai is still a work in progress, but some designs from Midjourney have already been successfully implemented.

Learn more about design on our Telegram channel. Subscribe and stay up to date with fresh ideas and trends in the world of design!

Neural networks are becoming an important tool for designers, allowing them to improve the quality of their work and speed up the creation process. These technologies can automate routine tasks, generate unique images, and offer innovative solutions for design projects. Modern neural networks analyze large volumes of data, which helps them generate creative ideas and inspiration. Using neural networks in design not only improves efficiency but also opens up new horizons for creativity. Designers can focus on more important aspects of their work, leaving the technical tasks to automated systems. The integration of neural networks into the design process is already changing the approach to creating visual content, and this trend will only gain popularity in the future.

• 6 Free Neural Networks for Creating Presentations
• 9 Neural Networks for Photo Processing
• Experiments with Neural Networks: 5 Fresh Cases
• 7 Neural Networks for Interior Design
• Kandinsky: How to Use Sber's Neural Network
• Midjourney: Immersion in Imaginary Worlds
• Stable Diffusion: What is this Neural Network and How to Use It