Topic: Why Does All AI Art Look Like That?
https://medium.com/@keithkisser/why-doe … 4e2a9e1c87
― Tennessee Williams
An independent forum site for members of the Serif software community
You are not logged in. Please login or register.
Automated signup has been disabled for these forums. If you wish to register, please send an email to admin[at]punster[dot]me stating your preferred username.
Embedded images: Click on the Add image to post link below the message box. Your photo will be uploaded to the postimage website and then shared here via a thumbnail link. Click on the thumbnail for a larger view.
Private messages: There is no pop-up alert for new PMs on these forums, so you need to look out for the PM link (near the top right-hand corner of any forum page) becoming bold.
Alfred's Serif Users' Forums → Art & Literature → Why Does All AI Art Look Like That?
https://medium.com/@keithkisser/why-doe … 4e2a9e1c87
An interesting article. I feel I have learned some things about art.
I noticed this within the text.
> Certainly, a few contemporary AI “artists” have never experienced any level of critique, until they posted their “work” on social media.
One thing that I have never done is purport that I produced the art generated by Bing Chat AI. I have always stated that the image was produced by Bing Chat AI. i have stated that I authored the prompt for that particular artwork.
I would be interested to know what readers, some of whom may be professional artists, think of four images referenced here by links to another thread in this forum, perhaps by reference to the comments about the art quality of AI images made in the article to which Alfred posted a link.
https://punster.me/serif/viewtopic.php?pid=3876#p3876
https://punster.me/serif/viewtopic.php?pid=3899#p3899
https://punster.me/serif/viewtopic.php?pid=3901#p3901
https://punster.me/serif/viewtopic.php?pid=4273#p4273
Clicking on an image enlarges it, clicking again enlarges it further,
William
Very good article, it seems to me that the only people who worship AI art are the ones who know least about the way that artists themselves judge the value of art.
What they all lack is proper composition, the values of balance, depth and inspiration. Similarlty to photographic values is supreme. Anatomical veracity comes a poor second.
If you want to know more, I have written a book on it. It is free to read online:
https://www.calameo.com/books/006503327ad682fc32541
John
Thank you for the link to your book, John. That is very kind of you.
William
I have got to part way through page 13 of John's book.
I have learned something interesting from that.
I had heard of the rule of thirds, as in 0.333 and 0.667, but it is really golden ratio at 0.382 and 0.618.
Ah!
So more centred than thirds.
So the gap in the middle is not 0.333 it is 0.236 so much smaller.
Ah!
William
I had heard of the rule of thirds, as in 0.333 and 0.667, but it is really golden ratio at 0.382 and 0.618.
…
So the gap in the middle is not 0.333 it is 0.236 so much smaller.
The 0.236 gap in the middle splits the 0.618, perhaps not surprisingly, according to the golden ratio: 0.236/0.382 is 0.618.
So what if one splits the 0.382 in golden ratio?
One way, the other way, or both ways?
Some more lines for the diagram?
And perhaps recursively?
I am thinking that a 1024 pixels by 1024 pixels png file with a transparent background and some of the golden ratio lines and thus points on it placed over (copies of, so not spoiling the originals!) some of the 1024 pixel by 1024 pixel AI generated images.
For example, is the mouth of the stegosaurus at such a point in this picture? Or not?
https://punster.me/serif/viewtopic.php?pid=3899#p3899
So wondering if the AI system has taken the golden ratio into account when composing the picture. That picture was requested to be in the style of Claude Monet.
William
So,
0.618 x 1024 = 632.832 so use 633
0.382 x 1024 = 391.168 so use 391
I shall now try to use Affinity Designer to make a transparent png with four red lines on it.
William
I produced the red lines as red filled zero width bordered rectangles 1024 pixels by 3 pixels, locating them using the centre of the 3 pixel width.
So, I consider it is worth trying with grid lines at one third and two thirds.
0.667 x 1024 = 683.008 so use 683
0.333 x 1024 = 340.992 so use 341
So, back to Affinity Designer.
William
Here is a 1024 pixel by 1024 pixel png graphic that I have made using Affinity Designer software.
The graphic has a transparent background, with the centre lines in cyan, upon which are the thirds lines in green, and the golden lines in red.
The intended purpose is to explore what composition rules may have been used in images produced using Bing Chat Ai.
Readers are welcome to use the graphic with images generated by other AI systems, but I do not know what size they are.
William
It may be that in order to "right click download" a copy that one will first need to click on it then click on the result so as to get the full size image to download.
The prompt to Bing Chat AI on Thursday 11 January 2024 for the picture of the lady feeding a stegosaurus was as follows.
Please produce an original painting in the style of Claude Monet of a lady in a long green dress feeding a stegosaurus in a garden where there is a pond with waterlilies in flower.
William
The grid lines added to the picture from
https://punster.me/serif/viewtopic.php?pid=3895#p3895
do not have the same result as with the stegosaurus picture, but there might be something.
The original picture is the one from the original prompt. I simply changed "an okapi" in a copy of the text to become "a stegosaurus" to explore whether Bing Chat AI would produce such a picture. In the event it produced a beautiful picture.
William
Moving on to page 14 of John's book.
Rabatment.
I wonder how rabatment would look on an A3 landscape picture, and on an A5 greetings card version of it.
Can rabatment and golden ratio be used in the same picture?
If so, golden ratio of the whole picture, or of one side or other,or both, of the rabatment?
William
Do the AI generated images presented in this thread have the faults with AI generated images that are purported in the article linked from the first post in this thread?
William
I have had a go at producing an original picture of my own using Affinity Designer so as to feature both rabatment for the end view of a house and golden ratio for the positioning of the horizon.
I do not know why I included the sun and why I positioned it where it is, it just seemed necessary to include it and the position just seemed right.
The perspective of the roadway just sort of happened, I used what Affinity Designer refers to as the Trapezoid tool and then used a -45 degree sheer and there is was, I just needed to then move it behind the building while keeping it in front of the grass and choose a colour for it.
The original is A3 plus bleed areas in case I decide to get an A3 print. The image here is one seventh scale both horizontally and vertically.
William
I am very much a beginner with producing art, so constructive criticism of my pictures is welcome.
William
You are doing very well William. However it took me a long time and a lot of looking at the work of skilled painters before my work started to show decent composition. Carry on reading about balance and leading the eye, also look at the work of Frank Webb, Toulouse Lautrec, Russell Flint, Degas and the Norwich School and Rowland Hilder. For more recent artists Ian Roberts, Charles Sovek, Mel Syabin, Gordon McKenzie.
Thet are all brilliant at composition, not forgetting Turner of course.
John
While recognizing that this thread has, due to me, drifted off-topic, but, well, with your forbearance please.
In July 2024 I tried a watercolour brush in the Affinity Designer software program, looking at how overlapping brush strokes look.
Well, rather than just arbitrary brush strokes I decided to try to sort of sketch a seascape of the tide coming in over a beach, thus sky, sea, sand.
Using only one colour for each, so three colours in total.
The result seems to have more than three colours,but that is due to the overlapping of the brush strokes.
I started with an A2 landscape canvas, then I added an A3 plus bleed areas rectangle in the centre.
First I did the sky, then the sand, then the sea. This order as I wanted to try to have the effect of the sand being visible through the water of the incoming tide where the leading part of the water was not very deep.
So most overlapping of brush strokes of sky upon sky, sand upon sand, sea upon sea, yet some sea upon sand.
As it happened, rather than just leave the result as an electronic image, I made a copy of the Affinity Designer file and in that copy I enlarged the guide rectangle a little to become double the size, both horizontally and vertically, of the size needed for A5 plus bleed areas, then exported that selection at exactly half size as a png (Portable Network Graphics) graphics file, using measurement in units of pixels for maximum accuracy. I bought a giclée print and framed it.
Until yesterday, I had been thinking that I should have put the horizon across the middle of the picture. But after reading (part thus far) of the book to which John kindly supplied a link, I now realize that it is good that I had not put the horizon across the middle.
So, bearing in mind that I am a beginner and how the picture came to exist, I am wondering what is good and what is not so good about its composition.
Is the horizon in the right place?
If not, where would it be better placed?
Is the sea and sand "quasi-horizon" where the sea meets the visible sand in the right place? I have sometimes thought it is too high up the picture, but it is an incoming tide, not the tide already in.
The illustration is approximately half in both directions of the A5 plus bleed areas png graphic, so around A7 in size.
William
Hello John
Thank you for your reply.
I have only just now noticed it. I had been writing my subsequent post for quite a while and producing the illustration graphic for it.
Best regards,
William
Alfred's Serif Users' Forums → Art & Literature → Why Does All AI Art Look Like That?
Powered by PunBB, supported by Informer Technologies, Inc.