Features of image generation using a neural network

It has become commonplace to ask the neural network questions, write posts with it, argue with it, wait for advice or chat about leisure time. But today we will talk about graphics and its characteristic strokes.

Odes are sung about the technical capabilities of AI, about the millions of photos used, about the principle of learning and about the process of transformation of a small or (large, we don’t know writing abilities) text paragraph, which, as if by a wave of a magic wand, turns this bowl with a schematic description into an incredible artistic masterpiece!

Put on your magic cloak and let's conquer artificial intelligence! In the absence of a broom, hat, witch's cauldron and other magical attributes, we suggest generating them with a neural network!

1 Definition of the task

Machine intelligence will help you in any activity! Have you dreamed of being a photographer and organizing photo shoots, but never reached an agreement with your first model via FTP? Do you draw by hand, but it’s a pity to stain your donated sketchbook with sketches? Do you blog about important topics and want to add useful information with pictures?

Oh no, you are a practical person and have nothing to do with creativity; you work in the field of construction, law, mechanical engineering, etc. In this case, you will need a business card, a photo to post on the website, or an illustration for a future banner. Choose a tool based on your needs and remember that some services are better suited for certain tasks.

2 Choose your style

To our joy, the neural network can create a real work using (almost) any stylization. An imitation of a painting painted in oil, tempera or watercolor, a still from a film or an advertising video, a creative photo shoot, even moving a world star to a nearby street. But only for illustration!

Let's imagine you didn't sleep at night watching the most famous anime in the world. Morning naturally comes and covers you with its inevitable reality, but you want to continue the Japanese banquet! It doesn’t matter, you can create similar graphics using the power of your mind, keyboard, mouse and enjoy again.

What can we say, you can convert your own images and select them as a profile photo in all accounts! (except for the worker, of course, there is a risk that they won’t recognize you at the corporate party if you don’t wear a Sailor Moon costume).

3 Reliable friend

You asked the AI ​​to portray the cutest kitten in the world and were unhappy with the result. It would seem that this baby is quite cute, but we know that there is room for improvement! What style is it made in, what color is it, what is around it? Is this a cartoon character or a realistic feline?

Accuracy

Correct queries and provide explanations in the text, as if you were describing the situation to someone in real life with specific nuances.

You should move from the general to the specific, move on to the little things gradually, after identifying the main thing. Initially, the style of the lamp in the background behind the chair does not matter if in the end you want to get a picture of the cat lying on it. It’s good that he doesn’t sharpen his claws on it!

Format

Specify the proportions and aspect ratio of your canvas, set specific dimensions and specify the file extension. If you forget, you will need to do these steps again.

Phased

A useful point for those who may not have thought about this. You can assemble the final picture like a puzzle!

Set the overall picture on the website, save the file, cut it piece by piece in a graphics editor, upload each of them to the service and add details. Repeat this process as many times as necessary until you get the result. The illustration will turn out perfect if you did not go beyond the edge of each of the parts and generated objects in the same style.

Tool

Midjourney, Dream Studio, Stable Diffusion, Kandinsky, DALL-E, whichever neural network you choose, first of all perceive it not as a rival in work or a competitor in the design field, but as a reliable friend who will make your daily routine easier and make your projects even cooler!

4 Life is in the little things

We asked the AI to create a realistic image, it worked, it’s ready! When you return to the material after a couple of weeks to admire your work, you are likely to notice that everything is not so smooth.

The risk of an error occurring due to carelessness is always present. To minimize it and if you want to get a realistic photograph similar to one taken with a professional lens, we recommend paying attention to the following points:

• Law of perspective.

The further away objects are from us, the smaller they appear and the more distorted and blurred they are on the horizon line. As a result, we can consider them in less detail. People who wear glasses may have adjusted them at this point, we are with you!

• The volume of the subject and details.

Each object, when illuminated, displays: light, shadow, penumbra, reflex. The reflex is a highlight by which we see the outlines of an object in the deepest shadow. If the item feels flat but should look different, rewrite the technical specifications on which the graphics are created.

• Proportions.

The window in the room is much larger than a box of matches. This doesn't mean the world should be measured in matchboxes, but be mindful of the height and width of the objects you create.

Did you know that the ratio of the head to body of an adult is approximately 1 to 8? And for a small child, 1 to 4? Does the size of the palm from the beginning to the tips of the fingers correlate with the height of the face from the chin to the hairline? To check the last question for yourself, if you don’t take our word for it, use an antiseptic or wash your hands in advance, we care about your health! You can notice the distortion of the listed relationships in the finished drawing that the neural network will produce.

• Clear boundaries.

When two objects merge as if they were doused with nougat in a package, but these are not two sticks of the famous sweet bar from different factories, then this is a reason to think and repeat the generation of the image.

• Interaction of light and shadow. Not good and evil!

Unlike works of art, where positive or negative characters oppose each other and only one of them takes the key position and wins, the light and shadow of an object must interact together and solely according to the laws of physics.

If an object is flooded with the sun or other light sources are directed at it from one side, then a shadow falls from the opposite side. When major defects have already been excluded, pay attention to this.

• Full face and profile of faces.

Uniting people with the surrounding space, distorting shapes and facial features will greatly affect the perception of the person who views the photo, so correct these defects.

• Look at your hands.

They may not generate properly. Pay attention to the number of fingers, incorrect anatomy, unnatural postures.

5 Emelya was lying on the stove, but you won’t succeed!

Why? You ask, if you were the hero of a fairy tale and had a stove, even now! But it won’t work!

If you think that artificial intelligence will do all the work for you, then you are mistaken. Yes, it will create an image, that’s right, it will be on the topic you need. But which one?

To accurately display all the necessary nuances, it is better to prepare the ground yourself. Develop your spatial imagination, you introduced the hero Emelya, so it turns out, congratulations!

For example, you intend to display references of an interior in which everything is arranged in a certain way. Which one? Visualize, you are the creator here. This is a living room, an office, a greenhouse, is it really a library named after Lenin? What's the atmosphere like there? Is this a picture in joyful warm colors or is it a dark, cold room that evokes gloom? What could be in this space? It’s unlikely that you have an artificial ecosystem with the rarest plant species in the middle of your office. But if so, we have a couple of questions!

The more accurately you describe what you want, the better the result you will get. Leave some percentage for the probability of error, yes, even AI can make mistakes, it is still learning!