Whether AI draw Gabdulla Tukai correctly

How neural network can help preserve the Tatar language. Part 2, image generation

Whether AI draw Gabdulla Tukai correctly
Tukay playing construction set?. Photo: предоставлено Райнуром Хасановым

Rainur Khasanov, the chairman of the World Forum of Tatar Youth and an IT specialist who uses artificial intelligence services daily to improve, simplify and accelerate work on various projects, continues to talk about his experience working with artificial intelligence, answering the question: how AI can help in the development of the Tatar language?

Creating images of consistent characters

“Generally speaking, there are many different neural networks for image generation," says Rainur Khasanov. “The most popular of them are the following: DALL-E (in conjunction with ChatGPT), Midjourney, LeonardoAI, Stable Diffusion model, as well as the Russian neural network Kandinsky. Today, we will look at some of them and their practical application for the Tatar language.”

When one uses AI to create illustrations for a story with one character, a problem arises. According to Khasanov, the same neural network draws different characters with identical text queries:

“To solve this problem, we had to make a lot of settings and describe the prompta in great detail. But recently, with the help of DALL-E 3 and the new GPT — Consistent Character GPT application, creating consistent characters has become much easier.

  1. Go to the GPT Store and type “Consistent Character GPT” in the search.
  2. Click on the “Start Chat” button.
  3. Click “Click here to start creating character designs”.
  4. Specify the gender of the character and the name (for example, in my case, let it be “Rainur”).
  5. Enter additional details: age, city, hairstyle, clothes, physique, skin colour, facial features.
  6. Be sure to add “Full body”.
  7. Choose an image style, such as Pixar Animation.

It is better to write prompts in English. The effect will be better, the expert explains. You can also ask the AI to change the character's pose (for example, when he walks or jumps), while preserving the attributes of the original image. It will take several attempts to achieve the desired result. It is advised to avoid terms like “face left” or “face right” so that there is no rotation of the image. The app is only available for paid ChatGPT Plus users.

This neural network can be used to generate various poses and expressions for games, animations, and comics. Naturally, everything will not work out perfectly, but it will already be possible to get a certain layout of the character with different poses and refine it in a professional application for artists.

Leonardo AI

Leonardo AI is a program that helps to create images and texts using artificial intelligence. It can draw pictures based on the description, edit photos and write texts, such as articles or promotional materials. This is convenient for designers, marketers and other creative people because it speeds up and simplifies their work.

In Russia, the service works without a VPN, offers 150 internal coins for generating images. The supply of coins is replenished every 13 hours. The program generates four images for 10-30 coins.

For example, we will ask it to generate an abstract character that is associated with Tatar culture. Does it look similar? I think you can answer the question yourself.

When generating images, the program does not understand Tatar, so prompts should be written in English. When generating, you can choose different models, sizes, styles, and more. Not so long ago, it became possible to generate a video from an image. We can try it sometime.

AI Comic Factory

There is an artificial intelligence model on the Hugging Face platform that generates comics. You can choose a style, write a story, choose the time of day and year. The AI will generate an image within 20 seconds. You can ask for something to be replaced, improved, and so on.

On Hugging Face, Galim Turumtaev, one of the activists of our Neurotatarlar community, created an AI model with which you can generate images of Tukai. Here is the link to the model.

You need to register on the service and find the Inference API Text-to-Image field, enter the request, starting with the words: “TUKAI man”. On the poster for this article, a young Tukai is playing construction set.

In the next article, we will consider other AI services.

Prepared by Radif Kashapov

Подписывайтесь на телеграм-канал, группу «ВКонтакте» и страницу в «Одноклассниках» «Реального времени». Ежедневные видео на Rutube, «Дзене» и Youtube.

Tatarstan