The following article complements the video published on the Beyond AI channel. We invite you to visit the channel, where you will find more information and inspiration on artificial intelligence. The channel's motto is: ‘Your guide to the dynamic world of AI.’

Creating graphics using artificial intelligence (AI) has become one of the most exciting applications of this technology. With the right tools, you can generate images that reflect your ideas, regardless of your artistic skills. In this article, you will learn which free tools to use, how to write effective queries (called prompts), and whether using AI to create images is legal and cost-effective.

How to create graphics using AI?

Hi, I'm Tymek, and I run projects involving the implementation of artificial intelligence tools in companies on a daily basis. Today, I'm going to shed some light on creating graphics with the help of AI tools. I'll show you how to use image generators in a practical way, preferably for free.

First, I'll point out that chat queries are called prompts. The terms ‘query’ and ‘prompt’ will be used interchangeably here.

1. Is it legal to create AI graphics?

Creating graphics using AI is a topic that became quite controversial at the beginning of the adventure with these tools. Some models used the works of many artists and were able to literally copy parts of their works, including even the signatures on the images they generated. At the end of 2024, the situation is completely different, and the largest AI software producers guarantee that their models learn from legal and ethical sources.

Yes, using AI-generated images is legal as long as we use proven tools. The giants really do not care about controversy, and they also have such data resources that they do not need to ‘steal’ any more.

2. What AI tools for image generation are available?

At present, there are several major players on the market, as well as many smaller models. We will focus on the largest, safest, and most stable ones.

If you are familiar with AI chat models such as ChatGPT, CoPilot, Gemini, Claude, and so on, you should know that image generation models are completely separate models. ChatGPT and Copilot use the DALL-E algorithm to generate images, while Gemini uses Imagen.

So we can guess that the images created by Imagen in Gemini will be different from those created by DALL·E in ChatGPT and Copilot. However, the issue goes a little deeper, because both ChatGPT and Copilot are wrapped in a so-called system prompt.

This is an additional set of descriptions, commands, and features that are added to each query we enter, which is why Copilot and ChatGPT will create different images despite using the same model to generate graphics.

We will focus on these three commonly available chats.

1. Gemini from Google
2. ChatGPT from OpenAI
3. Co-Pilot from Microsoftu

To use their free version, simply log in using your Google account details, for example.

3. Free AI graphics or pay?

Let's be honest, most of us just need to play around or create a few graphics once in a while. That's why it doesn't make sense to buy subscriptions for AI tools.

How do the free versions differ from the paid ones?

They have limits on the number of graphics you can create per day, and the counter resets after 24 hours, allowing you to create a few more graphics.

If you really want to play around more or need more graphics, then it's worth considering purchasing a monthly subscription – this usually costs between PLN 80 and PLN 100 for the largest models, although in this case I recommend MidJourney, where you'll pay around PLN 40 per month.

4. How to write a request for AI graphics creation?

This is the most important point of the program! The most common mistake is entering a query that is too general. For example, “create an image of a forest.” It is worth considering whether, if we gave such a task to a human, they would know exactly what we mean. That is why we must describe what we need very precisely. If the chatbot does not know this, it will do “as it sees fit,” which may not satisfy us. So where do we start?

4.1. Start with the big picture

“Create an image of a forest” is an appropriate generalization, and that is where you should start.

If we have an idea for a small house on the edge of a forest, which will only be a detail, we will still start the description by saying that it is a graphic of a forest, not a house near a forest. “On the edge of the forest stands a small house with a red roof.”

4.2. Add adjectives

Remember that you are talking to a chatbot, so it doesn't have to sound nice, and more specific descriptions will give better results.

“Create an image of a peaceful, misty, autumnal deciduous forest.”

4.3. Lighting descriptions

Lighting plays a very important role in the final effect.

“Create an image of an autumn deciduous forest illuminated by the setting sun casting long rays through the treetops.”

4.4. Style descriptions

The style can be photorealistic, it can be cubism, it can be oil painting, it can be Van Gogh style, and so on and so forth...

"Photorealistic style."

4.5. Composition descriptions

You can describe what should be in the image and from what angle the camera is “looking.”

“A photo taken from above the treetops. The horizon and the sun are visible in the upper part.”

4.6. Color descriptions

It's quite a simple thing to do, but it can make a huge difference. If you need graphics that match the visual identity of your company or project, you can always color the entire photo in specific colors or add details in the colors you want.

“Warm orange coloring with elements of green.”

4.7 Provide orientation and aspect ratio

For example: “Vertical photo, aspect ratio 3:2.”

Unfortunately, not every model responds well to such commands, so sometimes you have to ask again.

4.8. Avoid ambiguity

Remember that chat does not read minds, just like humans, so if you give it data that is only understandable to you, it will certainly not decipher it the way you want it to.

As we can see, the result of our query evolves in a very visible way, and although, objectively speaking, some of the intermediate results may seem more appealing to us than the final result, only the final result corresponds to what we need exactly.

5. Examples of good prompts

I will now show you five examples of well-written prompts and their effects.

1. Futuristic City at Night

“A view of a futuristic city at night, with tall glass skyscrapers illuminated by neon lights in shades of blue and purple. A busy street below, with fog hanging in the air, flying vehicles visible in the distance. The scene has a dynamic, cyberpunk vibe, and the lighting creates contrasting shadows.”

2. Magical Forest in Fairy Tale Style

“A fairytale forest at dawn, full of tall trees covered with soft moss, through which a delicate, golden mist permeates. Blue fireflies flicker on the ground, and in the background you can see a stream with crystal clear water. The whole scene has a mysterious, peaceful atmosphere, like an illustration from a book about elves.”

3. Portrait of a Victorian Lady

“A Victorian-style portrait depicting an elegant lady in a lace dress with a high collar. She has dark, curly hair pinned up in an intricate bun and wears delicate pearl jewelry. The background is dark and the lighting is soft, reminiscent of oil painting, giving the whole picture a classic, elegant look.”

4. Modern Scandinavian-Style Kitchen

"A bright Scandinavian-style kitchen with minimalist wooden finishes. In the center is a large kitchen island with a marble countertop, surrounded by white cabinets and shelves with green plants. The lighting is natural, streaming in through large windows, giving the interior a cozy and fresh look."

5. Epic Fantasy Scene with Warrior

"A warrior standing atop a mountain, holding a gleaming sword, with a picturesque view of the valley illuminated by the setting sun stretching out behind him. A dragon can be seen flying in the sky, and the clouds are illuminated with shades of orange and purple. The scene is dynamic and dramatic, like a frame from an epic fantasy film."

As you can see, the results are truly stunning and much better than those we get using very vague queries.

Summary

In this section, we summarize in brief what we have accomplished and learned in this episode.

To recap:

To begin with, use: Copilot, ChatGPT, Gemini

Models responsible for creating images include DALL-E and IMAGEN

When writing prompts, remember to:

• Writing from general to specific
• Providing adjectives
• Describing light
• Describing colors
• Indicating camera angle
• Describing atmosphere or overall effect
• Providing style
• Providing image format
• And avoiding ambiguity

An additional tip is that if you are not entirely sure how to write a chat query, you can ask for help from the chat:

> I want to send a request to the chat to create a forest graphic, but I'm not entirely sure how to do it or how to describe what I need. Ask me questions so that you can create the best prompt possible in the end.

Simple? Simple! I hope my instructions will help you achieve better results when working with chats. What else are you interested in?