Weaknesses of AI image generation

The biggest flaw in AI image generation software is that it completely ignores the basics and just starts coloring the picture itself. That's why it's so unstable and makes so many processing errors.
Normally, when drawing a picture, you decide on the composition, draw a rough sketch, make corrections as necessary, draw the line art, and finally color it. Don't just start drawing from scratch. If you don't have a solid foundation, you can't draw a picture properly.
So what are the basics of AI?
First, create the overall composition. Calculate the exact position, coordinates, and positional relationship of each element. Calculate the positional relationship between the subject and the background and foreground.
Next, generate and analyze the rough sketch, and verify and correct any processing and calculation errors.
Finally, generate the picture itself.
Of course, the quality of the learning data and the learning method are also very important. It is clear that the traditional learning method of showing a lot of images and photos to learn has already reached its limit. It goes without saying that AI image generation software cannot overcome all of the challenges it faces.
It is essential to create a system that can generate images, learn from corrected images, and recognize and correct mistakes.
Furthermore, a model like that used for face recognition and identification is also necessary, and it is necessary to learn the position, ratio, and shape of each part (element), such as two eyes lined up at the top, a nose in the center, and a mouth below that.
The current (conventional) learning and generation method is a system that is like copying a fake.
It is also important to add 3D models (wire layers) and motion data to the learning images to learn the shape and movement.
If AI images do not have a solid foundation, they cannot be expected to develop any further.
It is clear that a proper foundation is essential not only in image generation, but in all fields.
Automation will continue to advance in the future, and it cannot be stopped.
However, the current situation is that engineers and ordinary people are only focusing on processing speed and processing power, and neglecting basic learning.
Simply put, we are chasing results and completely forgetting the fundamentals, the origins.

https://note.com/poison_raika/n/n501ac31af2f1

<>