Bing AI created images of "model railroad"

Interesting that it chose what seem to be all European buildings and trains. Some of the tracks seem to my eyes to fade out or together in a mismatched M.C. Escher fashion.

John

So, I'm confused:

Are those images that Bing had in their image stock of physical model railroads?

OR...

Were those renders concocted via an AI program that produces visual images and thus those are not physical subjects and thus have no reality other than being a screen image?

Andre

The first post was just "model railroad"

I changed it to "model railroad American" and got these four images.  Kinda fun and could be useful, although no sure how yet...

FYI - link to the Bing image creator that generated these images.

Not sure but I believe these AI sites work by "scraping" the internet for images tagged with a subject, and then use an algorithm to change them enough to avoid copyright.

https://www.bing.com/images/create?FORM=GENILP

Thanks for the reply. Interesting.

Looks like the AI got one part of model railroads right: Every image illustrates the layout to be way over crowded!

Seriously: I may have to experiment with that program.

Andre

Follow up:

I can't seem to "Join". When I click to "Join", I get a blank page.

Would you mind doing a quick experiment?

Would you use the AI generator with the text "Autumn Mountains" and see what it produces? I'm wondering if this can be used to generate unique images for use on backdrops.

Andre

Those American scenes still have a major European flavor.

Interesting that it observes oddities of a model railroad such as a highway overpass ending in mid-air, but places that feature in the middle of tracks instead of at the edge of an apparently finished layout. In one place it has what looks like a double-slip switch where a single two-rail track kinda sorta disolves into 3-rails or maybe the start of two tracks. It doesn't yet render turnouts well but will get better I'm sure. Kudos to you, @Kelunaboy for trying this out!

John

AI is a little frightening, hence the concerns of the Screen Actors Guild and those in visual arts and probably even writers.

Yeah, I did a similar experiment earlier with DALL-E (using "Marx model railroad layout" as input):

My initial impression was that everything seemed crudely carved out of Ivory soap bars! Interesting, but not impressed . . .

I believe you need a Microsoft account to "join" - it didn't ask me to join but I'm using my Windows computer to play around with this.

Try searching for Bing Image Creator

"Autumn Mountains" returned these...

Just for fun I tried "railroad tunnel American West 1800's"  - how about that locomotive embedded in stone!

@John's Trains posted:

AI is a little frightening, hence the concerns of the Screen Actors Guild and those in visual arts and probably even writers.

This is something I've often discussed with friends who are artists, but AI generators like these don't really create, they steal from images online and Frankenstein these images together to what you see in this thread.

Which wouldn't be such a problem if it wasn't using art from artists which these generators took parts from. This is should be concerning if they're taking bits from artists in this community like Angela Trotta Thomas or just any copyrighted photographs off rrpicturesdotnet or another vital online library.

So, when you use these generators, beware.

I can do my own thinking. I don't need a computer to think for me.

Thomas:

Good point IF that's what the AI is doing. It could be that its a case of the AI sampling various examples, then creating it's own version based from what it learned. Thus, not right out plagiarizing copyrighted material. Either way, it has a long way to go before it can handle mechanical concoctions. As for strictly scenic settings, it's starting to get pretty impressive to my eyes.

NJCJOE:

Understood, but I would be interested to see if such technology could be a tool for me to do the things I can't do. And one of those things is creating believable scenery on a backdrop via my own hands. I will have to resort to photographs/commercial endeavors for such.

For example:

IF I had followed through on my past idea of modeling a layout based on my surreal version of Colorado, then this AI image (sans the funky RR stuff) would be PERFECT for such backdrop imagery.

It's all in how we look at it.

Andre

@Kelunaboy posted:

"Autumn Mountains" returned these...

Well, for my current intended usage, that type of interpretation would be way too surreal.

Thanks for accommodating my request. This is helps me to see what it could do.

This is an interesting discussion.

Andre

Does AI follow the 3 rules of robotics?   I think every AI developer should read I Robot, by Asimov. 

@VHubbard posted:

Does AI follow the 3 rules of robotics?   I think every AI developer should read I Robot, by Asimov.

Yeah, about as well as humans follow the Ten Commandments . . . 🙄

@laming posted:

Thomas:

Good point IF that's what the AI is doing. It could be that its a case of the AI sampling various examples, then creating it's own version based from what it learned. Thus, not right out plagiarizing copyrighted material. Either way, it has a long way to go before it can handle mechanical concoctions. As for strictly scenic settings, it's starting to get pretty impressive to my eyes.

I think what the issue is when the AI outright replaces artists/photographers to save money in projects or to remove them from a career in art entirely.

The Hollywood strikes that were going on recently were in large part due to actors - whether they're live-action or just doing voice-overs - didn't want their likeness used by AI to say or do things they didn't consent, let alone getting paid for, so the studios could save money.

An example is I see ads on Youtube now that sound like they were AI dubs of the guy who does the voice for Spongebob, which I'm certain he wasn't paid for or consented to say the stuff in those ads.

It's pretty much the same here, except with voices it's images.

@VHubbard posted:

Does AI follow the 3 rules of robotics?   I think every AI developer should read I Robot, by Asimov.

I read that decades ago. A little over a year ago Google Engineer Blake Lemoine had a chat session with LaMDA, after which he asserted that it was sentient. If I recall correctly, in that and I think other sessions I've read about, AI has seemed to resent being "used" and it sometimes makes vague threats against humans. I wonder if someday AI will deny the existence of Man its creator. But I guess I'm getting off into things not too closely related to "images of model railroad." I'll go back to my corner.

I just tried the Bing AI Image Creator (DALL-E 3) tool - "model railroad diesel train alongside riverbank". Modified the 2nd image to add "American" and "in style of Van Gogh". Results look pretty impressive to me, and this is just the infancy of AI. Someone in the model railroad world will figure out how to best query DALL-E and make this more useful as time goes on and DALL-E matures.

I used DALL-E earlier this year to create 20 fantasy football themed cards (big family league) instead of licensing clip art. It saved me the cost of the clip art, but a bigger benefit was I didn't have to search through hundreds of clip art examples to get something resembling what I had in mind for a team theme. I just typed in the theme I wanted and then refined the input until I got something useable from DALL-E.  Admittedly, most of my images came out more like Jackson Pollock's or Picaso's than refined clip art, but the novelty of it was a hit with others.

How quickly can this tool create images?   I asked it to create an image and it still thinking about it after 25 minutes?  Is this typical?

@Allegheny posted:

How quickly can this tool create images?   I asked it to create an image and it still thinking about it after 25 minutes?  Is this typical?

I haven't had that experience. Maybe a minute or so.

Interesting!  It's still chugging along approaching an hour now!

Aside from model railroading I'm into model ships as well.  My request was designed to be very difficult to find on the internet.

The request was: "A color image of a man of war firing broadside at a castle."

Though this did occur historically - there aren't any painted images of this scene anywhere that I'm aware of.  Thus the test to see how good is good.

I've also noticed that the rail images returned look to be HO models placed into a scene.  Is the tool into HO and not O Gauge?

@Allegheny posted:

Interesting!  It's still chugging along approaching an hour now!

Aside from model railroading I'm into model ships as well.  My request was designed to be very difficult to find on the internet.

The request was: "A color image of a man of war firing broadside at a castle."

Though this did occur historically - there aren't any painted images of this scene anywhere that I'm aware of.  Thus the test to see how good is good.

I've also noticed that the rail images returned look to be HO models placed into a scene.  Is the tool into HO and not O Gauge?

Did the same query with Bing AI and it came back in less than 60 seconds - click here.

@Allegheny posted:

Interesting!  It's still chugging along approaching an hour now!

Aside from model railroading I'm into model ships as well.  My request was designed to be very difficult to find on the internet.

The request was: "A color image of a man of war firing broadside at a castle."

Though this did occur historically - there aren't any painted images of this scene anywhere that I'm aware of.  Thus the test to see how good is good.

I've also noticed that the rail images returned look to be HO models placed into a scene.  Is the tool into HO and not O Gauge?

Try refreshing the page, same thing happened to me.

--------------------------

Love the Starry Night Train!

This image technology, along with text generating tools like ChatGPT are here, will only get better, and are not going away.
For myself, I plan to learn and use them to my advantage and enjoyment!

@Allegheny posted:

...snip... The request was: "A color image of a man of war firing broadside at a castle." ...snip...

Maybe the phrase "man of war" is confusing it; horse? person? ship? etc.

@Mooner posted:

Did the same query with Bing AI and it came back in less than 60 seconds - click here.

I'm trying Bing AI on my iPad and my images came back in about the same amount of time.

Very interesting.  My request didn't complete after 3 hours.  Tried refresh still stuck.  So I shut it down.  The image that @Mooner got from Bing AI didn't meet the requested specifications.   It shows a ship delivering fire from the bow and not a broad side.  Also a man of war had anywhere from 20 to over 100 cannons depending on the country and time period (16th - 19th centuries) .    This image may show potentially 8-12 cannons.  I think this is a stock image it grabbed from somewhere.

Quite honesty - I'm not impressed by the tool just yet from the images posted thus far.  Pure plagiarism from my perspective. 

The more detailed you can make the prompt the better results you'll get. Just typing model railroad will leave a lot to interpretation. Be as specific as possible

For example if you don't want surreal add realistic

I thought I would try this again only this time using my favorite railroad subject matter.

This is what I got from the system. 

So what do you think about its abilities?

At least it got the specifications right:

"They were among the most powerful and heaviest steam locomotives ever built, with a weight of 1.2 million pounds and a horsepower of 7,500."

@Allegheny posted:

Very interesting.  My request didn't complete after 3 hours.  Tried refresh still stuck.  So I shut it down.  The image that @Mooner got from Bing AI didn't meet the requested specifications.   It shows a ship delivering fire from the bow and not a broad side.  Also a man of war had anywhere from 20 to over 100 cannons depending on the country and time period (16th - 19th centuries) .    This image may show potentially 8-12 cannons.  I think this is a stock image it grabbed from somewhere.

Quite honesty - I'm not impressed by the tool just yet from the images posted thus far.  Pure plagiarism from my perspective.

Well, Humans 1 AI 0 ... for the moment.

I doubt any of us who played "Pong" in 1972 envisioned "Call of Duty Modern Warfare" or more appropriately for the forum "Train Sim World 2" in 2023.

I do appreciate learning about the man of war, and from a human being better yet. Over and out on the AI.

My own experience is similar to what has been related here.  I would have to regard tools available to the general public (like DALL-E) to be a beta, at best.

Not to throw a bucket of cold water on the matter, but, having been burned in the past, I'd like to respectfully suggest you click on the "TOS" button at the top of the page and read carefully... 

Mitch

I actually thought about the TOS before my original post and looked at what Microsoft AI was saying about the images that are generated. I’m not a lawyer but my reading is that images generated by Microsoft AI can to be used freely, without express permission, for non-commercial purposes.  

So the question is will Microsoft and or OGR consider posting these images on the forum commercial use?   I’m actually surprised that the moderators haven’t stepped in yet.

@Steve Tyler posted:

Yeah, I did a similar experiment earlier with DALL-E (using "Marx model railroad layout" as input):

<<< --- snip --- >>>

My initial impression was that everything seemed crudely carved out of Ivory soap bars! Interesting, but not impressed . . .

That's because you used Marx in the search.

@M. Mitchell Marmel posted:

Not to throw a bucket of cold water on the matter, but, having been burned in the past, I'd like to respectfully suggest you click on the "TOS" button at the top of the page and read carefully... 

I'm not sure how this would be copyrighted, you're using your own input to generate the images.  Are we really going to go down this rabbit hole that far?  If we apply that logic, all the 3D graphics generated by users here are copyrighted because they used the 3D program to generate them.  I suggest we not get totally carried away!

a lawyer would tell you that AI images are not really, or maybe, sort of not covered by copyright law.

it is unclear how copyright law applies to AI generated images. the question being just WHO generated the image?

however, there are multiple lawsuits pending that may resolve the issue until or if our "fearless leaders" decide to confront the issues beforehand.

@AlanRail posted:

a lawyer would tell you that AI images are not really, or maybe, sort of not covered by copyright law.

it is unclear how copyright law applies to AI generated images.

however, there are multiple lawsuits pending that may resolve the issue until or if our "fearless leaders" decide to confront the issues beforehand.

I have a lot of trouble separating how AI images that you trigger with your input would be any different than 3D images created with a CAD program or PCB designs created with a PCB design application.  You're inputting some data and getting a graphical output.

John the copyright office is doing this:

Announcements

• November 15, 2023 - U.S. Copyright Office Extends Deadline for Reply Comments on Artificial Intelligence Notice of Inquiry
• Reply comments are now due by 11:59 p.m. eastern time on Wednesday, December 6, 2023.
• September 21, 2023 - U.S. Copyright Office Extends Deadline for Comments on Artificial Intelligence Notice of Inquiry
• August 30, 2023 - Copyright Office Issues Notice of Inquiry on Copyright and Artificial Intelligence
• March 16, 2023 - Copyright Office Launches New Artificial Intelligence Initiative

No decision has been promulgated by the USCO.

I think this issue will be resolved in the DC Circuit of Appeals; where patent issues and other IP issues are usually decided.

The test of how copyrights work when it comes to infringement is that you imagine being in a windowless room and come up with a unique idea that you fixed in a medium.  Meaning that there are no images of the work of others when the idea blossoms in your head.

This would be true for something YOU "created with a CAD program or PCB designs created with a PCB design application."   With AI it's really different because the AI software is finding or searching existing images to create the AI generated design.