United3dArtists Archive

FHembree · Post by **FHembree** » 13 Sep 2024, 23:01

I saw a large number of these images and animations on X after the Trump/Harris debate and I can't help but believe these were totally created by some type of A.I. software. Am I correct in this assumption?

Finis · Post by **Finis** » 14 Sep 2024, 02:45

AI image generators are getting better and people are getting better at prompting them. Some things about these images indicate AI but others contradict that. Just a guess: They began as AI images then improved or corrected by a human or it is AI and human made combined. Someone is very fast with photo editing if these have no AI content.

I'll use the term "AI" since it is commonly used to refer to programs like MidJourney.

MOTORCYCLE

AI evidence: distorted aircraft, unidentifiable cloth object on his back, cat has a thumb able to wrap around handle bar, cat is not holding the rifle, gas tank has a distortion. The background has the AI look.

Not AI evidence: Hands and paws are good. AI's tend to mess up hands. Motorcycle grip, firearm, reflector on gas tank are detailed or functionally correct/believable to a level I haven't seen from AI. Left hand is present and on the left grip. AI's get confused when arms or legs are behind things.

Unknown factors: I haven't seen motion blur in an AI image before. Translucent cat ears.

ACTION CATS

AI evidence: Faces have identical patterns. Rifle butts don't look functional or correct. Backpacks and side things are unidentifiable. Gold chest emblems don't look like anything.

Not AI evidence: Trigger discipline, their fingers are not on the triggers. They have a finger. AI's tend to mess up hands.

Unknown factors: Water. Translucent ears.

.

bitkar · Post by **bitkar** » 14 Sep 2024, 12:42

these are clearly AI... must admit that those guns look quite OK... its a bit of a problem with what i have at home.

FHembree · Post by **FHembree** » 15 Sep 2024, 05:54

Yeah, you are right. I thought I would experiment and I searched and found that Microsoft was a free site where you can experiment and play around with a software called Designer.
You just type in a description and it renders 4 images in just a few seconds. I typed in a few words for a modern house of precast concrete in the woods next to a river and etc
One of the four results is attached. I admit its a lot of fun to experiemnt with this software.

RAYMAN · Post by **RAYMAN** » 15 Sep 2024, 07:13

Problem with this kind of software is that it kills our planet !
It uses tons of servers and international computational power and makes our
planet hotter then it should be !
We have down under here in Europe with rain every day and floods !
In Austria Neather-austria is in alarm at the moment with the Danube flooding
moast part of the country passing on slowly to the Chech Republik !
I would love to know how much power consumption all the servers
use to run all the Data banks to make this kind of stuff plus its much more fun
to make by hand and say its your creation ! Not my world anymore !
Did I say i miss the good old 3D and photo world with art and knowledge !!!
Not using AI !!!
Its a no go for myself !!! One exception maybe scaling and noise reduction !

bitkar · Post by **bitkar** » 15 Sep 2024, 11:46

i dont think so... it does not produce more heat than PC gaming or even hours of doing graphics and test rendering. There are NPUs for that, that are a bit more effective. And creation of one image take only few seconds or even less on a big server. I run stable diffusion locally on my PC using my GPU and its not more energy hungry than when I play games.
Its more a bitcoin problem, that produces heat for no reason.

RAYMAN · Post by **RAYMAN** » 15 Sep 2024, 15:16

its running over how many thousand servers.......

dont forget rendering texturing finding storing...sometime rigging.....
would be interesting to find out !

Finis · Post by **Finis** » 15 Sep 2024, 15:58

That picture looks nice FHembree. The colors, object placement, and lighting of "AI" images are often pleasant.

Skilled prompting can produce good results. The images have an AI style or look that is remarkably similar across all the the different generators. One factor is probably that there are many generators but only a few methods implemented. Lately some are producing images so photo realistic that the AI look is not present or is subdued.

Creativity and imagination such as some humans have surely appears because it is in the human made source material the program was trained with. That's probably true of the compositional value as well.

bitkar · Post by **bitkar** » 15 Sep 2024, 22:26

with open source AI generators like Stable Diffusion you can download many different models. It does not stop there. I see it as a totally new branch of graphics design. There are more than just the AI model, you can continue, with LORAs, ControlNet and other modules/plugins which give you control over the result (even style - photoreal, painterly, manga style, etc.). Its not all just that AI will generate something and you are not involved. For professionals its the oposite. Then you can modify the image using an Inpainting techniques and others.
IT IS A HUGE WORLD. Trully a rabbit hole.
And yes, there are servers, but they can creat many many images. My PC (not a server) can make 512x512 image in just 2 seconds. So the GPU consumes about 240 W for 2s to make an image. Try that with 3D and other software, i can tell you that you will consume more energy to make the same image. Ofc, If I do something more complicated and want to have control over details I need to generate more than once.

FHembree · Post by **FHembree** » 16 Sep 2024, 04:25

I won't upload anymore of these images as I know it doesn't require any real creative artistic skills, but I am truly amazed that anyone could write a software program that can interpret written sentences and then create images from that. I know a little about programming and it boggles my mind. Yeah I know that it understand basic key words such as "house" and that would mean a box type element and you specify the materials desired and that houses will have windows. Even knowing that little about it is amazing and it's only gonna get more sophisticated over time. I can see it becoming a tool for designers to saving time by being able to provide precise written description and horizontal and vertical dimensions instead of physically drawing floor plans and elevations to create a rendering.
I was impressed how just adding word snow storm it rendered the image with snow.

Below is the description I used for the rendered image

Modern precast concrete house over a falling stream in the woods in the style of frank loyd wright with round corners and vertical elements of stone, horizontal elements are precast in a snow storm.

bitkar · Post by **bitkar** » 16 Sep 2024, 09:34

the same prompt in SD:
(using base model, if i used some architecture model, it could have been better maybe)

Finis · Post by **Finis** » 16 Sep 2024, 19:45

Pretty pictures again! Too cold.

FHembree's latest one combines Wright's signature cantilevered terraces with curves and other prompt requirements well. It is not just the parts stuck together but creates a harmonious or cohesive design. Bitkar's made a good building but omitted the curves and has only a hint of Wright style. Since they use the same prompt this shows the differing focus or ability of different programs. Of course additional tuning, guidance, options etc. could improve either.

"AI" pictures even at current development are good for ideas or inspiration. Also for quick things like art for the walls of an architectural rendering. The linguistic part, following a written prompt, may be more impressive than the image making part.

Here is an experiment. Give that same prompt again, the one you both used, with the same options or settings to the same programs. See if they make the same pictures.

Steinie · Post by **Steinie** » 16 Sep 2024, 20:49

Well to see what good ole AI could do I asked it to create a Wright home with palm trees and skunks.
Now I want to live there!!!

bitkar · Post by **bitkar** » 17 Sep 2024, 16:41

usually those painting looking picutres are from DALL-E, i think.

Finis: here is my AI with the same prompt. Ofc its very similar, but never the same. Used the same settings. It creates random seed everytime. I can generate the same thing if I leave the same seed. Everytime iam amazed of how much it understands light, reflections and such.

FHembree · Post by **FHembree** » 17 Sep 2024, 20:16

No, you never get the same results. It generates 4 images each time you say generate image. I uploaded or saved the one I liked best of the 4. If you run the programme again with out any changes to the text you get 4 new different images.
The more times you run it with out changing the text the results get stranger and stranger.

FHembree · Post by **FHembree** » 17 Sep 2024, 20:24

Here's second run.

Finis · Post by **Finis** » 17 Sep 2024, 22:21

Similar but different images could be good. Similar wall art, inspiration for a housing area where variation within a style is wanted, many uses.

Steinie are there pigs near one house?

Strange images can be interesting or funny.

RAYMAN · Post by **RAYMAN** » 18 Sep 2024, 07:47

interesting I never prompted for the skunks and they apeared in my neiborhood without invitation !!!

hashtag : I prompted good reliable 5 star uptown belair residancy.....!

United3dArtists Archive

A.I. images

A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images

Re: A.I. images