Taking another look at AI and seeing how wrong it can get it

You all know I like playing around with AI and the results can be good and mixed. However, it can also be frustrating, and you may end up with a monumental failure.  A couple of months ago, I did a post about how you can get reasonable results, you can see the post here.

When I did that post, I tried some other things and they just didn’t work, so I thought for fun today we could look at what I asked for and what I got.

In the last post, I tried to make images of things that were famous or landmarks in Australia. Though to get those images, I also got some real stinkers. I realised when I picked what I thought were real landmarks in my city of Melbourne, AI in Photoshop had no idea. I thought, really, is Melbourne that unknown?

Then I thought, let’s do a post with the dodgy AI images.

The first one I thought of doing was the Webb Bridge. It is a famous bridge for Melbourne, so here is what I asked, and you can see what I got.

Generate an image of the Webb Bridge going over the Yarra River in Melbourne at sunset.

None of the images look anything like the Webb Bridge. They are so different that they could be a bridge anywhere. Certainly not the bridge I know.

Here is a photo of the bridge that I have taken.

photo challenge 1 - Bridge

Next, I thought, why not try the Westgate Bridge? It is possibly the biggest one here in Melbourne and was built in the 70’s. Also quite famous for collapsing when it wasn’t done properly.

Here is what I asked.

Generate an image of the Westgate Bridge with a container ship going under it on an overcast day.

This is what I got. Again, nothing like the Westgate Bridge.  The river is not that wide there, and if I’m honest, these remind me of images that I saw when I was in New York.

Here is a photo of the actual bridge, and you can see that maybe the first image above is the closest to it, but really not that close.

So I decided after these bridges to try another big bridge here, the Bolte Bridge. So here is what I asked.

Create an image of the Bolte Bridge with the city of Melbourne in the background.

I can’t get photos of the Bolte with the city behind it. You can’t stop on the bridge to take photos, and there is nowhere on the other side to do it. It seemed that AI might be the only way to get that.

I was wrong. I mean, there is a bridge and a city on the other side, but the bridge doesn’t look anything like the Bolte Bridge. The above bridges look more like pedestrian bridges, but the Bolte is part of a freeway so there are many lanes of traffice that go over it.

Here is what it looks like from Docklands.

As you can see, it is very different from how AI depicts it.

I don’t think AI is that great yet. I’m not worried about it too much. I know I will always prefer to use my own images, though I like using some AI to remove things or things like that.

As time goes on, I do think it is important to keep up to date with what it is doing and what it is capable of.  I know I would never use it to create my images, where would the fun in that be? I would only ever create images to see what it can do. In this experiment, it failed big time. The images don’t even look like photos.

I have another post like this planned, but it will be a lot more fun, so find out next week.

You might be interested in …

25 Comments

    1. I think you are right, I find it interesting to keep up to date with what it can do. I don’t think we need to fear it, but we also don’t have to accept it. Thank you Geoff.

  1. Yours are so far superior I don’t know what to say so I will borrow from Margaret21 “cheapest of cheap greetings cards with wildly over-optimistic skies”. They make me queasy and they are showing up in all the image searches at times more than the real images.

    1. Thank you Elizabeth, I know what you mean about the searches and on places like YouTube, how can you believe what you are actually seeing anymore.

    1. It is interesting and getting better and better all the time. Thank you Esther, I think mine are more representative of the prompts, that’s for sure. I like those ones too.

  2. And you were quite specific with your requests of AI. Good to know that we are still smarter.

    1. Yeah I was Lois, I think it probably has more to do with asking for Australian landmarks that probably are only famous for here. I am sure if you asked it to create an image of say the Grand Canyon it would do a decent job. I think it depends on how many images of the thing you want to create are out there. Thank you Lois.

  3. I’ve had some monumental failures too using AI. Another reason I always declare AI when I do. One day you’re bound to get caught posting a three-eyed goldfish.

    1. Yeah, I think we all get them. I do as well, though sometimes I wonder if I put up some photos that I thought AI did okay with would people be able to tell. I do have some funny ones, sort of, coming up next week. Thanks Mike.

  4. Not only are they apparently inaccurate. They’re horrible. Like the cheapest of cheap greetings cards with wildly over-optimistic skies. AI clearly has a long way to go yet.

    1. Sometimes I wonder if you just hate AI so much that you just can’t see the benefits of it at all. These are not accurate, but they aren’t necessarily bad images. I think the ones of the ships going under the bridge are quite nice, not what I was looking for, but not too bad. It is one of those things where you have made that decision and you aren’t looking at them objectively. I might be wrong, but others don’t seem to have the same view that you do Margaret. So I just wonder if your prejudices against it make you not look critically at it. I was watching some videos on YouTube and they looked real, but the story behind them didn’t. So I did some research and found out they were AI, but boy they had me fooled.

    2. You definitely make some fair points Leanne, but in my eyes, it’s the artificial Glorious Technicolor of the images I dislike so. I’ve seen AI images I like, but these are not they! AI is here to stay, and I accept that (even though it’s putting two family members out of work) but so often, as here, the images it produces are brash and unreal looking. I prefer more subtlety!

  5. Except for using as an editing tool, I don’t like AI. I’m wondering if you could give it a picture and say create a bridge like this, what would it do? If that were possible?

    1. I agree, I use it a lot for editing, but sometimes I like to see what it can do. I don’t know if you can do that Anne, I think you can add to an image, I don’t know that I care enough, lol. Thank you Anne.

  6. I am not a big fan of AI. I asked someone to check over my artist statement and make any changes that they thought it needed. When I read it I was sure that the person used AI because it used phrases and words that did not represent me and it sounded like an AI response. Does that make sense?

    1. I think it is a bit hit and miss. I like it for some things, but others not so much. I would agree with you about that, how can AI write a statement that would sound like you. I use Grammarly on computer and sometimes it askes me to change something and if I did it wouldn’t make any sense. Absolutely makes sense to me Mama.

Comments are closed.

Discover more from LEANNE COLE

Subscribe now to keep reading and get access to the full archive.

Continue reading