My top neural networks for a designer
Hello everyone, my name is Denis, I am the team lead of the design team at DLS, and also a reviewer at Yandex Praktikum courses “Interface designer” And “Mobile Application Designer”. In this article I want to share a selection of neural networks for communication and graphic designers who work with vector and raster graphics. And at the end I will tell you which template helps me make a good promt.
Neural networks are trained on large amounts of data and are designed to solve specialized tasks, helping people perform various tasks more efficiently.
Here are some examples of using neural networks in design:
Creating visual content — neural networks are capable of generating images, layouts and even entire design concepts based on given parameters
Inspiration and new ideas — can offer fresh ideas and visual solutions that will inspire designers
Trend Analysis – are able to analyze current trends in design and predict future changes and popular styles
Solving routine problems – background removal, image correction
Writing texts – drawing up work plans, describing images, creating catchy or target texts
Using neural networks greatly simplifies my work when analyzing trends and solutions of other designers. And also when I need to make a work plan, write text, prepare images for creatives, develop a concept for corporate graphics and logos, when solving routine tasks. I believe that the ability to use neural networks in work will soon become mandatory in the work of designers. It's time to adopt useful tools, simplify your work and increase your professional weight.
13 neural networks for a designer
1. Adobe Sensei
A set of neural networks integrated into Adobe products (Photoshop, Illustrator, etc.) helps automate repetitive tasks, improve images, and create design solutions. Adobe products have been greatly improved thanks to neural networks and have become even more convenient to work with. Now they can more accurately highlight elements, generate necessary objects in photos, and replace and extend backgrounds.
If you haven't updated your software yet, now is the time to do it. However, all this pleasure requires a subscription, and for residents of Russia, this may be a problem. If you have a foreign card, let's try. The cost of a subscription varies, depending on the set of programs included in the package.
The neural network generates images based on text descriptions. In the editor, you can change image settings, style, and orientation. It copes well with generating photos and illustrations. The program is conditionally free: 25 credits are provided per month, which can be spent on generating content. Then you need a subscription from $6 per month.
3. Spline
Here appeared the function of generating three-dimensional images by text request. It is possible to generate quite suitable models for your creatives. But the solution also requires a foreign card to pay for the service. The subscription cost is $24 per month.
4. Midjourney
With Midjourney you can generate photos, illustrations, abstract backgrounds, 3D images, logo concepts. I told you how to generate a logo in Midjourney earlier in in a separate article.
Quite a multifunctional solution that works directly in Discord as a chat. The cost of the service is $10 per month. To pay for it, you also need a foreign card. You can use a special Telegram bot that allows you to use Midjourney and the Pro version of ChatGPT for 500 ₽ per month. I will tell you about the bot later in the article.
5. ChatGPT
I generate fishy texts in GPT, ask it to briefly retell the contents of long texts and highlight the main points. The service has a bunch of extensions built on the GPT model and tailored for special purposes. For example, DALL E helps generate pictures in the chat. Using the service is conditionally free. The Pro version will cost $ 20 per month, you will need a foreign card to pay.
6. YaGPT
If you don't want to bother with payment and only need texts, YaGPT will help with this. Just follow the link ya.rulaunch Alice and ask her a question in the chat.
7. Racraft
My favorite. This is a cool neural network for generating raster and vector graphics, its interface is intuitive, similar to Figma or Spline. You can generate photos, backgrounds for your creatives, 3D icons and even craft a picture based on a reference.
Recraft also has a function for removing and replacing the background, replacing the selected area or drawing the necessary elements. The application has an active community, and you can use images created by the community in your work.
Recraft is shareware, you can generate anything in unlimited quantities, but to increase the priority of requests and get exclusive rights to use images, you need to pay a subscription – $ 20 per month.
It can generate images based on a text request, as well as answer questions and write texts. It is quite suitable for generating raster graphics, fish texts or article templates. The service is free.
9. YandexART in the app «Masterpiece»
Presented by the mobile application “Shedevroom”. In the neural network, you can generate raster images for your creatives, absolutely free.
10. Gervin
Domestic content generation combine. The service can create images, texts for posts on social networks in different styles, rewrite and translate texts. It will be especially useful for communication designers. The service works on a subscription basis from 280 ₽ for 10,000 tokens.
eleven. Gigapixel 7 and Photo AI by TapazLabs
Gigapixel 7 — allows you to upscale an image up to 600%. It will be useful for those who work with printed products. Upscaling is also useful for displaying images on screens with high pixel density. The application is paid — $100, 12 months of updates are included.
Photo AI — allows you to improve photos, remove noise, adjust lighting. The application will be useful for those who work a lot with raster graphics and photos. Cost — $200, 12 months of updates included.
12. RemoveBG — quick background removal
If you need to quickly remove the background from an image, RemoveBG will help you with this. The service has a trial version, you can remove the background from five photos for free. Then you need a subscription or a new account.
13. Bot in Telegram «ChatGPT + Midjourney» – a cocktail of neural networks
Two top neural networks in one place. In the chat, you can generate images and work with texts. The bot has a free version, with some limitations, and a paid version – 500 ₽ per month for both. I use this bot all the time: convenient and profitable.
How to work with neural networks
The neural networks described above can be conditionally divided into three types:
Networks that work like an app where you can press buttons and select the optimal settings.
Neurons that process text queries.
Hybrids.
If everything is relatively clear with the first, then with the second type of neural networks it is necessary to interact in a certain way, composing clear prompts (requests) in order to obtain the required result.
What is a promt and how to compose it correctly
Promt — a text query recognized by a neural network to generate content or perform a certain set of actions. Depending on the capabilities of neural networks, prompts can be made in Russian or any other language.
Sometimes it is difficult for a neural network to grasp the exact intentions of a user, as a result of which its response may not match expectations. Therefore, it is important that all requests are formulated clearly, without ambiguity, and contain all the necessary information for the correct answer.
I suggest you try the following template:
General instructions — sets the general style of response and role behavior of the model: “You are a poet. You always answer with a four-line verse.”
User request — a question or instruction that will be executed by the model during the generation of the answer: “What is the sky?”
Context — additional information required by the model to prepare an answer: “Form an answer using the word “azure.”
Response format — description of the answer format or an example: “Prepare your answer in the form of a poem.”
The full prompt (request) will sound like this: “You are a poet. You always answer with a four-line verse. What is the sky? Form an answer using the word “azure”. Prepare the answer as a poem.”
And here is what ChatGPT responded to our request:
“The azure sky is a boundless element,
There are stars, light and the play of clouds.
Piercing the heights, they soar as if in the dreams of the messiah,
And the heavenly bark beckons into the distance”
And here is what YaGPT answered us:
“The sky is an endless azure,
What attracts us with its beauty.
It seems to whisper: “Don’t be sad!”
And gives us peace”
If the neural network can generate both text and images, then to create an image, you need to write “draw a picture” or “create an image” at the beginning of the request. Then supplement the prompt with a request compiled according to the template.
Sometimes a situation arises when the neural network does not know the answer or part of the answer to a question. In this case, it can answer: “I do not know the answer to this question” or “I do not have information to answer.”
The model can also give a confident but incorrect answer. This is called a neural network hallucination. Such hallucinations can contain incorrect facts, logical and other errors that differ from the expected results.
The reasons for the model's hallucinations can be different:
• Lack of context – the neural network does not have enough information to give a correct answer
• Incorrect wording of a question or instruction
• Settings for internal response generation parameters
• Insufficient amount of data used to train the model
• No fact-checking mechanism before generating a response
Therefore, all content created by neural networks must be checked before being used in work.
Earlier I talked about the third type of neural networks – combined. They are made in the format of full-fledged graphic editors and additionally contain consoles for entering prompts, for example, Recraft, Adobe Firefly, Masterpiece. The above-described rules for composing prompts are also applicable to them.
It is up to you to decide which neural networks to choose for your work. Obviously, the time has come to study these useful tools and use them in your work to simplify your routine. In the near future, mastering the skills of using neural networks in a designer's work will be as mandatory as the ability to use Photoshop or Figma. That is why Yandex Practicum has created a separate course “AI tools for designers”There, in 4 weeks, they will teach you how to solve design problems using neural networks.
Thank you for your attention to the article! Tell us in the comments whether you use AI in your work. If yes, which neural networks do you like more.
