Kitabı oku: «Incredible artificial intelligence Easy Diffusion 3.0. The Middle and High School Guide», sayfa 3
Step 7. Which model to choose?
Which model should you choose to create the most realistic image?
In fact, this is a very important and difficult question. To find the answer to this, you and I must determine for ourselves what goal we are setting and what image we want to get as a result of our work.
Our main goal in the book is to get acquainted with the work of the Easy Diffusion program and learn how to create beautiful images dedicated to the topic of Artificial Intelligence, consolidating the acquired knowledge with numerous examples that are in each chapter of this book.
All images that appear in this book were created using Easy Diffusion version 2.5 and version 3.0 using the following models:
– Realistic Vision версия 5.1
(https://civitai.com/models/4201?modelVersionId=114367);
(https://civitai.com/models/4201/realistic-vision-v51);
– DreamShaper версия 8.0
(https://civitai.com/models/4384?modelVersionId=109123);
(https://civitai.com/models/4384/dreamshaper);
– Color Fusion версия 1.0
(https://civitai.com/models/20617/color-fusion);
– SDXL Yamer’s Realism! 2.0
(https://civitai.com/models/136669/sdxl-yamers-realism-realisticanime3d);
– Midjourney версия 4.0
(https://huggingface.co/prompthero/openjourney/blob/main/mdjrny-v4.ckpt);
(https://huggingface.co/prompthero/openjourney/blob/main/mdjrny-v4.ckpt);
– Openjourney версия 4.0 (модель обучена на рисунках Midjourney)
(https://civitai.com/models/86?modelVersionId=27392);
– Stable Diffusion версия 1.5
(installs with Easy Diffusion).
Below you can see a comparison of the results of four different models:
There are a large number of sites (information resources) on the Internet where developers post their trained models. I prefer to use the site civitai.com, which has not only models, but also various examples of work, as well as related requests.
Please note that in this book I focus specifically on the topic of Artificial Intelligence. As you’ll notice a little later, most of my queries include the keyword «cyborg.» This is determined primarily by my vision of what Artificial Intelligence should be. My views on this topic were formed a long time ago on the basis of many science fiction novels and stories that I have read in my life. These literary works were written by such outstanding writers as: Stanislav Lem, Boris and Arkady Strugatsky, Isaac Asimov, Vladimir Vasiliev, Sergei Lukyanenko, Harry Harrison, Edmond Hamilton, Robert Sheckley, George Lucas, Andre Norton and many others. In most of them, a robot, an intelligent machine and Artificial Intelligence are, in essence, some kind of single creature from the future, which is most often presented in the form of a person. Therefore, for me, no cyborg can exist without Artificial Intelligence, which replaces the human mind, and that is why in my images you will see many different futuristic humanoid robots.
Working with Easy Diffusion
I suggest you experiment a little and create your own unique images based on just one request, but using different models.
To complete this exercise, you need to download the following models from civitai.com:
– DreamShaper версия 8.0
(https://civitai.com/models/4384/dreamshaper);
– SDXL Yamer’s Realism! версия 2.0
(https://civitai.com/models/136669/sdxl-yamers-realism-realisticanime3d);
– ProtoVision XL
(https://civitai.com/models/125703/protovision-xl-high-fidelity-3d-photorealism-anime-hyperrealism-no-refiner-needed);
– Art Universe
(https://civitai.com/models/123313/art-universe);
– Realistic Vision версия 5.0
(https://civitai.com/models/4201/realistic-vision-v51).
Very often developers change the name or link to the model. Nevertheless, you can always find the model we need on the website civitai.com by typing its name in the search bar.
Now let’s do the following:
– Prompt: «Nouvelle AI, new Artificial Intelligence, cyborg face, an enchanted technological portal from the future, journey into the healing power of nature, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail.»
– As a model, we choose any of those available to you. I will give examples of models that are most relevant to the theme of the book.
– We will leave the remaining Easy Diffusion parameters unchanged.
Please note that if you are using Easy Diffusion 3.0, you must use the latest versions of all models to create images, otherwise Easy Diffusion may throw an error and stop working. To avoid problems with Easy Diffusion, follow one of the links and download the latest version of the model you need. For example, I encountered that the Realistic Vision model version 4.0 works fine in Easy Diffusion version 2.5, but does not work in Diffusion version 3.0.
Click the «Make image» button.
I got the following results for different models, but the same query:
Model DreamShaper версия 7.0
Model DreamShaper версия 8.0
Model SDXL Yamer’s Realism! версия 2.0
Model ProtoVision XL
Model Art Universe
Model Realistic Vision версия 5.0
Please note that if you have Easy Diffusion version 3.0 installed, you must be using the latest versions of all models.
For the exercises presented in all subsequent chapters of this book, also use the most recent versions of the models.
Step 8. Uninstall or reinstall Easy Diffusion
Uninstalling Easy Diffusion is even easier than installing it.
However, before deleting it, save the «Stable Diffusion UI» folder in any convenient location. It stores copies of all your work (images and request files).
The folder is located in one of the following paths:
– C:\EasyDiffusion\profile\Stable Diffusion UI,
– D:\EasyDiffusion\profile\Stable Diffusion UI.
Now you can safely delete the entire folder with the Easy Diffusion program by right-clicking on the folder and selecting the «Delete» command.
If for some reason Easy Diffusion stops working for you, there is nothing to worry about. To solve this problem, you need to take just two simple steps:
• Completely remove the old version of the program, keeping the necessary files. To restart a computer.
• Install a new version of Easy Diffusion by first downloading a new distribution of the program from the website: https://github.com/easydiffusion/easydiffusion.
Detailed instructions are presented in the second chapter of the book.
Sometimes it happens that after updating the version of Easy Diffusion from version 2.5 to version 3.0, images may no longer be created. This can be solved by removing old models and installing new ones. Detailed instructions are also presented in the second chapter of the book.
As we can see, in just eight simple steps we learned how to install, configure and uninstall the program, as well as create beautiful images.
Next, we will take a closer look at the functionality of Easy Diffusion and learn how to create our own digital works of art.
Terms and Definitions of Incredible Artificial Intelligence
Term 1. Artificial Intelligence
Today, scientists, engineers, marketers, programmers and other specialists from different areas of the economy use completely different definitions of the term «Artificial Intelligence». This is due, first of all, to the fact that Artificial Intelligence is widely used in various types of human activity in various sectors of the economy. For example, Artificial Intelligence is used in education, medicine, public administration, finance, industry, automotive industry, astronautics and many other areas. Each of these areas has its own unique features and specialization of its application.
The very concept of Artificial Intelligence appeared a long time ago. The term was first coined by scientist John McCarthy in 19562,3.
So what is Artificial Intelligence?
Artificial Intelligence (AI) is, first of all, a computer program written by a person (programmer, engineer, scientist or other specialist) that most often imitates the behavior or mental activity of a person or some other living creature living on our planet.
The presence and very existence of Artificial Intelligence is inextricably and closely related to computers, new information, computing and network technologies.
In the case when the operation of Artificial Intelligence requires large computing powers of modern supercomputers and a set of various auxiliary software products, it is spoken of as a combination of technical and technological solutions, the main task of which is to ensure a high level of «intelligence» of such a system.
Today there are a large number of different types of Artificial Intelligence. If all types are considered from the point of view of information processing methods, then it can be noted that the most popular among engineers and programmers is a method called machine learning.
In 1959, Arthur Lee Samuel, who is called the «pioneer of Artificial Intelligence», was given the following definition: «Machine learning is a field of study that gives computers the ability to learn without explicit programming.»4,5,6.
The figure below shows machine learning, which is a component of Artificial Intelligence:
Accordingly, Artificial Intelligence is trained on various types of data, using the machine learning method and very often its variation – the deep machine learning method. The data on which Artificial Intelligence is trained is called big data. The greater the volume of information and its diversity, the easier it is for Artificial Intelligence to find patterns and the more accurate the result will be.
There are many types of machine learning which are listed in the figure below:
There are a great many types of machine learning, and a separate book could be dedicated to them. If you later decide to continue your acquaintance with the world of incredible Artificial Intelligence and its technologies, I strongly recommend that you (in high school or college) understand the principles of operation of machine learning algorithms. This will allow you to understand how Artificial Intelligence «thinks» and will allow you to formulate for yourself the principles of explainability and trust in the results of its work.
Big data is a term for sets of digital data. The large size of data and its complexity require significant computing power of computers and special software Tools for their analysis and presentation. Big data includes arrays of numerical data, images, audio and video files7.
Scientists and engineers, improving various computing and information technologies, strive to create self-learning and autonomous Artificial Intelligence systems, which should approach the intellectual and functional capabilities of humans in their capabilities.
All Artificial Intelligence systems can be divided into two large groups according to the degree of autonomy:
– Systems of subordinate Artificial Intelligence – systems with the help of which some intellectual functions are automated and special intellectual tasks that were previously solved only by humans are solved. Such systems are subordinate to man and his target functions and do not contradict him. In our case, Easy Diffusion is a slave Artificial Intelligence system.
– Autonomous Artificial Intelligence systems that independently model the work of the human brain (memory, thinking, creativity, emotions, will, freedom of choice, search for new knowledge, etc.) and independently make optimal decisions. Ideally, such systems should be ethical and comply with the four basic laws of robotics.
From the point of view of scientists, Artificial Intelligence is a computer system based on a complex of scientific and engineering knowledge, as well as technologies for creating intelligent machines, programs, services and applications, simulating the thought processes of humans or living beings, capable of perceiving information and learning with a certain degree of autonomy and make decisions based on analysis of large data sets8,9.
From a software engineer’s perspective, Artificial Intelligence is a field of computer science that combines computing technology with robust data sets to develop computer programs to perform tasks that can mimic human intelligence – discovering meaning, generalizing and drawing conclusions, identifying relationships, and learning from taking into account the accumulated experience10,11,12.
Artificial Intelligence technologies are technologies based on the use of Artificial Intelligence, including computer vision, natural language processing, speech recognition and synthesis, intelligent decision support and advanced Artificial Intelligence methods.
Visually, the technologies that are used to create various Artificial Intelligence systems can be represented as follows:
As we can see, the definition of the term «Artificial Intelligence» is inextricably linked with the definition of the term «Artificial Intelligence system».
An Artificial Intelligence system is a computer program that is the implementation of new information processing technologies for the purpose of searching, analyzing and synthesizing data from the world around us in order to obtain new knowledge about it and solve various vital problems based on it. An Artificial Intelligence system includes models and algorithms that provide its ability to learn and present (visualize) new data in the form of text, numbers, audio, video or images13.
It should be noted that there are quite a lot of classifications of Artificial Intelligence systems. The most commonly used classification is:
Thus, we can say that today Artificial Intelligence combines a combination of machine learning (deep learning and reinforcement learning), machine thinking (planning, scheduling, knowledge representation, search and optimization), computing technologies, big data technologies, supercomputers, as well as robotics (control, perception, sensors, actuators and integration of all of them into cyber-physical systems)14.
In this book, we will introduce you to various definitions of Artificial Intelligence, and you will also be able to see with your own eyes the visualizations created based on these definitions in the Easy Diffusion Artificial Intelligence system.
Working with Easy Diffusion
After reading this chapter and having the basic knowledge of Easy Diffusion, let’s try to create a new image based on the knowledge about Artificial Intelligence that you gained from this chapter.
To strengthen our skills in working with Easy Diffusion, let’s do a simple exercise and try to create an image of Artificial Intelligence.
To do this, perform the following steps:
• Construct your own queries based on the definitions of Artificial Intelligence above.
For example, we wrote the following:
a. «A boy hugs an Artificial Intelligence robot against the background of a light sky, the camera colorfully focuses on them.»
b. «Machine learning, intelligence, mind, perfect composition, cinematic lighting, crouching near the camera.»
c. «Machine learning, intelligence, mind, cinematic lighting, robot, cyborg.»
• We chose three models as models: «SDXL Yamer’s Realism!», «Realistic Vision» and «DreamShaper».
Click the «Make image» button.
We got the following results:
Text-to-image AI software: Easy Diffusion
Prompt: Artificial Intelligence that will change the world, white picture
Model: SDXL Yamer’s Realism!
Image Author: Alexander Chesalov
Text-to-image AI software: Easy Diffusion
Prompt: Machine learning, intelligence, mind, perfect composition, cinematic lighting, crouching near the camera
Model: DreamShaper
Image Author: Diana Chesalova
Text-to-image AI software: Easy Diffusion
Prompt: Machine learning, intelligence, mind, cinematic lighting, robot, cyborg
Model: Realistic Vision
Image Author: Vsevolod Chesalov
Term 2. Narrow Artificial Intelligence
Most Artificial Intelligence systems that people use today are varieties of «narrow» (one might even say «narrowly focused» or «highly specialized») Artificial Intelligence systems.
Artificial Narrow Intelligence (ANI) is Artificial Intelligence that is trained and able to effectively perform only certain highly specialized tasks. Weak Artificial Intelligence is the most common application of modern Artificial Intelligence technologies, which primarily include so-called «machine learning» and «deep machine learning» (which we discussed in the previous chapter)15.
Modern narrow Artificial Intelligence systems are programmed to perform one simple or complex task at a time, extracting information from a specific set of data. All their work most often comes down to the constant performance of similar tasks in order to obtain the best results in a narrow field of activity.
Weak Artificial Intelligence includes: image and facial recognition systems, chat bots, conversational assistants, self-driving cars, recommendation and analytical information systems, and so on.
The familiar Yandex Alice, Apple Siri, Amazon Alexa or the IBM Watson supercomputer are also, oddly enough, classified as narrow or weak Artificial Intelligence.
All versions of the ChatGPT program, created by OpenAI, which is capable of writing not only posts on the Internet, but also creating its own literary works, also belongs to narrow Artificial Intelligence.
The Easy Diffusion program we study in this book is a weak and highly specialized Artificial Intelligence system. We set the task for this system to create a new unique image for us from our description or from another image, and Easy Diffusion solves this problem very well.
So why is Artificial Intelligence still called weak?
Weak Artificial Intelligence, no matter how «smart» or «strong» it may seem, today cannot compare with the capabilities and potential of human intelligence. He has no will and is not capable of creativity. Weak Artificial Intelligence is not able to function independently, is not capable of self-service, self-development, self-improvement, reproduction and interaction with other Artificial Intelligence systems, as people do.
Despite all this, this type of Artificial Intelligence has many undeniable advantages, which lie in the fact that it is capable of performing specific highly specialized tasks very quickly, efficiently and accurately, sometimes even better than a person himself. Where a person may be very tired, Artificial Intelligence can help him do the work on his own, and Artificial Intelligence can do this for a long time. The collaboration of humans and Artificial Intelligence greatly affects the productivity and efficiency of the work performed. The described process is often called «automation of routine tasks,» which greatly simplifies our daily lives.
The use of weak Artificial Intelligence gives us more time for self-development, relaxation and achieving new goals.
Working with Easy Diffusion
To consolidate our work skills, let’s do a simple exercise and try to create an image of highly specialized or weak Artificial Intelligence.
To do this, perform the following steps:
– Create your own queries based on the definition above. For example, we wrote the following:
– «Artificial Narrow Intelligence, stupid, dumb robot trying to solve puzzle, cyberpunk.»
– «Artificial Narrow Intelligence, a large iron computer carries a cup of tea to a man, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail.»
– We chose DreamShaper as a model.
– Set the «Image Quality» parameter to 75.
Click the «Make image» button.
We got the following results:
Text-to-image AI software: Easy Diffusion
Prompt: Artificial Narrow Intelligence, stupid, dumb robot trying to solve puzzle, cyberpunk
Model: DreamShaper
Image Author: Diana Chesalova
Text-to-image AI software: Easy Diffusion
Prompt: Artificial Narrow Intelligence, a large iron computer carries a cup of tea to a man, photorealistic, perfect composition, cinematic shot, intricate details, hyper detail
Отрицательный запрос (Negative Prompt): black background
Model: DreamShaper
Inference Steps: 50
Качество изображения (Image Quality): 75
Image Author: Alexander Chesalov
We wanted to get Easy Diffusion’s answer to the question of what it thinks of weak or narrow Artificial Intelligence. We got some pretty funny pictures.
Try creating your own image using the keyword: «Artificial Narrow Intelligence.»
