I have lots of topic to write in this blog. First of all, it’s time to write ‘make a picture with A.I.’. As usual, this post is published little late considering about 80% of materials in this post was ready in May 1st. It takes about 40 days to fill remain 20%.
Many people, including me, expect the subject which A.I. replace human’s work as writing and repetitive research, while art is not easy to replace. In real situation, repetitive research will not be replaced soon. As I work in research institute, I have seen that people are using their brain very creatively to work less. And there are so many impressive news title (to get more attention). Rather, drawing and music can be made mechanically. Now, the quality of A.I. pictures appoaches useful level and many people are interested in using A.I. I encounter the A.I. drawing with Bing image creator.
The example below the ‘아이디어 탐색'(In English, ‘Explore Ideas’) are fully available and many object can be drawn.
After knowing A.I. drawing was opened to general public, I started to use ChatGPT with paid statues to draw. (The real computer in 80486 generation is much more fancy than above images. How stupid chatGPT is..) If you are sensitive, the main icon of this blog was get then by chatGPT drawing.
And I tried to draw ‘a beautiful woman’. (I think it is very usual to draw something beautiful.) The drawing ability of chatGPT is very high, I get so high quality of picture after simple instruction.
Then, the DALL-E module in chatGPT had many limitations. (For example, instruction can be related to make output with sexual is blocked.) I understand that there is a risk of misused. But I have some regrets. I can’t order to draw a girl with ‘short skirt’ while sometimes DALL-E draw it without instruction. After then, I met ‘stable diffusion’. It had less ability to under understand natural language compare to chatGPT. But the drawing ability is quite great. All pictures below are drawn by stable diffusion.
Hmm.. My instruction was ‘Yellow hair girl in the library’. Above picture by default setting of stable diffusion doesn’t deserve above praises
Hooray~ It is also made by stable diffusion with similar keywords. This picture is focused on the cartoon style.
And this picture is focused on cute Asian girl style, while the hair color is not yellow but brown for reality.
Wow.. It is real ‘yellow hair girl in the library’. I can’t find any fault. lol
I think stable diffusion is very nice program, since it can make various pictures by similar keywords. (See above pictures.) The product is strongly influenced by the ‘study’. The program draw similar pictures compared to what it studied before. So the picture style can be changed by instructor. I recently get to know how to study and its algorithm. Human think it is ‘draw the picture’ but computer just performs ‘numerical correction’ of image file. So the program doesn’t need to study ‘art’ itself.
Moreover, the studied contents are open in internet. I don’t need to study. In stable diffusion, it is called as ‘check-point model’. The above pictures are drawn with downloaded check-point models. I got to know how to download check-point model and draw. Recently I know how to improve the keyword and how to study. Now I can draw with advanced keywords and study my own style, so I can say ‘I can draw with stable diffusion program’ now.
The stable diffusion is a ‘program’ installed in my PC. So it has lots of freedom – it can draw adult only picture both of violent scene and sexual scene. I have seen so many pictures couldn’t be drawn in Dall-E, I recognize some channel notice that the pictures are drawn by stable diffusion. And I confirm that stable diffusion can use ‘nudity’ keyword. I won’t open the result, of course.
While Dall-E in chatGPT is internet based A.I. program (It means the calculation server in somewhere in the world) and get a result after short time, stable diffusion is a program installed the user’s computer (Actually, it is one kind of python library. so the user have to install python.) and the operation is performed in my computer. I realized why the A.I. technology is related to electricity consumption. When I request stable diffusion to draw, my CPU usage became 100% and fan start to run loudly, and the picture is out after ten to twenty minutes. I worry that my computer will be broken soon if I am interested in A.I. drawing. I will draw sometimes.
If I want a picture in the blog post, I will draw it by A.I. How wonderful modern technology is!
