When Reason Sleeps, Ghosts Awaken

Alone in a suspicious laboratory a young man discovers new reality

How AI-propelled images are being created?

You have probably already heard that AI will take artists’ jobs, kill our creativity, and who knows what else. Allow me to introduce myself: I am a master of arts in visual communication (specializing in photography), and I will prove to you that, on the contrary, AI can greatly enhance the quality of everything you want to create.
It will not take my life away, it will improve it.

However, there is a catch: if you can create anything you can imagine, your images will only be as good as your imagination.

And that, my friends, is the most fair rule in art, in my humble opinion. The better your mind, the better you can use words, and the more you devote yourself, the more you will improve and excel.

Now, regarding those who argue against AI, especially in relation to media created with its help, the issue with most of them is not that they are not entirely right, they are not even wrong!

People criticize AI without actually trying it themselves, wanting to ban images created with the help of AI as if no knowledge is required to create such artworks. They overlook the countless variables, organized by certain rules, that any AI-using artist must learn.

Think you can create those images by simply writing a simple prompt in an online AI image creator? Well, go ahead and give it a shot.

Can AI generate hundreds of good-enough random images with a few clicks? Yes and no. Yes, with the right prompt, you can generate many good, nice, and even beautiful or artistic pieces. However, as we accept that, the expectations from viewers and customers rise, so we end up in a situation where someone (AI) has just turned the table around.

How these images were created?

I utilize Stable Diffusion technology.
Essentially, I have graphic “models” that have been trained on a variety of images, such as manga, horror themes, fantasy, and more. These models are created and shared by talented AI graphic pioneers. One notable open model is called RealisticVision2.0, and there are other popular models like Midjourney. You may also meet yourself with the expression ‘Checkpoint‘.

These models/checkpoints are essentially files that can be downloaded onto a computer and used with the assistance of a graphic user interface to create images.

Each model requires a prompt or description to generate the desired image. The prompt serves as an input that guides the model’s output. There are extensive tutorials available, including on platforms like YouTube, that explain the rules and techniques for creating effective prompts. As AI technology continues to advance rapidly, there is always something new to learn and explore in order to maximize the potential of these models.

Additionally, each model or checkpoint can be combined with three different types of modifiers: Textual Inversion, LORA, and LyCoris. These modifiers can be seen as mini-models with varying levels of complexity, where the less complex ones are less controllable. Textual inversion allows the insertion of specific personas, such as mini-models of celebrities, while LORA can manipulate the overall image style to a controlled degree.

For example, the images in the gallery you see were created using a Horror-type LORA modifier that I obtained from CIVITAI
As a base model, I used midjourney.

While the current resolution and size of the generated images may not be on par with expectations, it’s important to note that improvements are being made.
The limitations in resolution and size are partly due to the VRAM (Video Random Access Memory) constraints, which are not extensive even on modern graphic cards. However, advancements in technology will likely lead to gradual improvements in these areas over time.

But the most important part is writing a good prompt. (See great tutorial on that! -> https://stable-diffusion-art.com/prompt-guide/ )

This means that someone who may not excel at drawing but has a background in humanities like history, archaeology, or art history can still create compelling images. Why? It all depends on what ideas are in your head. If you can imagine anything and there are no limitations, what would you create? How well can you describe it? Is it a scene from a movie or a historical event you would like to depict?

I have also experimented with GPT-Chat, creating specific rules and a semi-mini language that we can use as a template to create highly detailed images. It is a valuable AI tool that can greatly assist in image creation. However, working with it requires careful management and it is best utilized in collaboration with humans.

Personally, I have developed a sort of partnership relationship with GPT, which I find quite satisfactory. I like to believe that we create these images together, and this collaboration between AI and human holds its own value.

To this day, I have not been able to depict any of the historical events on my list, such as the Battle of Grunwald (known as Bitwa pod Grunwaldem) depicted in Jan Matejko’s famous masterpiece. The Battle of Agincourt and the lesser-known Battle of Crecy are also on that list.

However, I’m not sure if this is an AI limitation or simply a result of my current basic skills. Creating realistic, photography-style images with a focus on historical accuracy seems to be beyond my artistic and IT capabilities. I have also noticed a scarcity of detailed battle scenes on the internet, suggesting that it may be a challenging task.

What factors influence image creation?

IF YOU THINK THAT CREATING AI-GENERATED IMAGES IS EASY, JUST TAKE A LOOK AT THIS LIST OF VARIABLES THAT WERE USED TO CREATE THE PICTURES DISPLAYED IN GALLERIES.

Creating AI-generated images involves a variety of factors and variables that influence the final outcome. Some of the key elements that can affect image creation include:

• Model Selection: Selecting the appropriate AI model or checkpoint is vital in influencing the style and quality of the generated images. Each model has its own unique capabilities and training data, leading to diverse outputs.
  However, this is only at a beginner level. For more personalized effects, professionals often combine up to three models together. Additionally, it is possible to train your own mini-model, further expanding creative possibilities.
• Prompt Design: Crafting an effective prompt is essential in guiding the AI model to generate desired images. The prompt should provide clear instructions, context, and specifications for the image generation process.
• Modifiers and Techniques: Employing modifiers and techniques like Textual Inversion, LORA, or LyCoris can further manipulate the style, content, or specific aspects of the generated images. These tools allow for fine-tuning and customization.
• Training Data: The dataset used to train the AI model influences the visual references and patterns it learns from. Diverse and extensive training data can enhance the model’s ability to generate a wide range of images.
• User Skill and Expertise: The user’s artistic vision, creativity, and technical understanding contribute significantly to the quality of the final images. Experience and proficiency in utilizing AI tools and techniques can further enhance the outcomes
• Further graphic skills in applications like those from the Adobe family or similar free alternatives can be highly beneficial and are often necessary at a professional level. These tools provide additional capabilities and enhance the overall creative process.
  
  
  Indeed, it is important to note that creating great images is not solely dependent on the elements mentioned. There are numerous free internet sites available that offer AI-powered picture generation, providing accessible tools for users to explore their creativity. I will dedicate a separate post to delve into this topic further.
  
  
  At the end, I will also present to you my dramatic failure, a rather ridiculous experiment in depicting the Battle of Grunwald. I tried at least 5 different models, with rather surprising results.
  
  For those who are not very familiar with that event, the Battle of Grunwald is a famous medieval battle that took place in 1410. It was fought between the Teutonic Knights and an alliance of the Kingdom of Poland and the Grand Duchy of Lithuania. The battle resulted in a decisive victory for the Polish-Lithuanian forces and is considered a significant event in the history of Central and Eastern Europe.
  
  The result is… well you can judge yourself.