How does it work? How do you create such images? What can’t be done? What is the actual quality of these images? Is it going against the current, or can I use it for my own purposes? Should I struggle against it (well, it will be easier with the sun) or adapt to it? (Here’s a little hint: humans are the most powerful species thanks to their best skill in what? Adaptation.)

Does that mean that years of studying (artistic disciplines, painting, graphic design) are useless and have gone to waste? Should you regret your decision?

Absolutely not. In fact, your knowledge and understanding of image creation are more valuable than ever. With the advent of AI, you can now leverage your skills to create images that were previously unimaginable. All your skills and experience will still be required to achieve a professional or master level, and AI image generation does not exclude that. The better artist you were, the more you can squeeze from cooperation with AI.

That is me – AI cautious enthusiast – after 3 months of experience with new reality

AI is a powerful tool, but it relies on the instructions and descriptions we provide. It can generate images based on our input, but it cannot replace the understanding and expertise of photographers, painters, and artists. Concepts like lighting, techniques to convey emotions, and the art of creating impressions still hold immense importance. AI is a means to enhance and expand our creative possibilities, but it cannot replace the human touch.

So, instead of rendering your years of study irrelevant, AI empowers you to explore new horizons and push the boundaries of your artistic expression. Your skills, combined with the capabilities of AI, can lead to the creation of images that were once beyond your wildest dreams. Embrace this new technology as a tool to amplify your creativity and unlock new artistic possibilities.Such topics will be featured on the website. Such topics will be featured on the website.

I will show you: what, how, where?

What: I will show you what the main app for creating such images is, as well as the addons that are worth checking out.

How: I will demonstrate how to install new apps and, more importantly, how to create specific images, whether they are realistic, surreal, paintings, or any other style you desire.

Where: I will provide you with a list of websites that are worth checking out. These sites are ideal for creating AI-generated pictures or finding informative tutorials on “how to…” topics.

I will guide you through the process of creating each of the showcased images, including how to describe and generate them. It’s also valuable to explore these techniques on your own, as new addons, such as new models, embeddings, or LORAs (and extension models for ControlNet), are continuously being developed, enhancing the capabilities at various levels every day.

An attentive observer would certainly notice that the presented images have certain flaws. I intentionally selected imperfections, or let’s say, “candidates for further editing,” rather than the final presentation. But perhaps this way, it will somewhat reassure people who are afraid that AI will suddenly take their place.

It will take some more time, and here’s why it will be so: like any tool, AI-generated images also have their weaknesses, which means it’s not yet the all-powerful killer that some portray it to be.

Of course, almost every problem has already been identified, or research teams are working on solutions. However, there are recurring observations that the AI community widely reports, and here is their list:

• AI-generated images are a bit like a lottery; you never know exactly what will be generated.
• Therefore, generating numerous versions of the same idea is necessary.
• Even images that appear decent may have several minor flaws, such as consistently poorly generated hands, fingers, and related details.
• Serious errors in generating faces can occur. While there is an option to fix and regenerate faces, it can clash with the process of tile-based upscaling.
• The dream effect—I call it somnambulism—often causes details to blend into shapes that may resemble something at first glance, but upon closer inspection, we realize it’s an abstract pattern instead of, for example, fingers on a hand. Often AI images look like stills from David Lynch’s movies.
• Generality—AI can generate a tank, but it may not be able to generate a specific tank like the Abrams. It can produce something resembling the tank, but many details won’t match the original. (However, specialized mini-models called LORA already exist, capable of generating something specific, like a French soldier from World War I. But someone dedicated trained such a model on actual photos or paintings.)
• It’s hard for me to imagine that AI will suddenly become a killer of packshot creators because it cannot generate something it hasn’t encountered before, like a specific designer version of clothing or any other product that has yet to be introduced to the market.(Arguable – I know there is a at least one pro tutorial on YT how to do a specific bottle photo, but I do not think that AI workflow can beat professional photographer, at least not yet.)
• Weak recognition of quantities, numbers, letters, signs, etc.—AI cannot write (although there may already be remedies for this).
• The issue of size—current models mainly rely on the Stable Diffusion 1.5 model trained on 512×512 px images. Stable Diffusion has already introduced the 2.1 model for 768px, but the new model did not generate much enthusiasm among users. Surprisingly, better results can still be achieved using the older 1.5 model and its derivatives, supplemented by upscaling techniques. *
• The legendary idea that everything will magically happen with just one click can be dismissed. While not every detail needs to be perfect for internet publication purposes, if you want to do professional work using AI, the creation process takes much longer.

*The SD-XL version 0.9 of the model trained on 1024px images has already been announced. Whether it will prove to be better will soon become apparent. (#update) is published and available for everyone.

If you want to try your hand at it, check out this post. It provides detailed instructions on two ways to create AI-generated graphics using a website and a more professional and complementary approach, which involves building your own UI to control Stable Diffusion.

Please READ THIS TUTORIAL FOR AI GRAPHIC ON HOW TO BECOME AI USER