By now we’re all-too familiar with AI-generated images of cars, among other things. Some of them are quite artistic, and we’ve had our fun with them here in the past (links below) to create What-If scenarios. But in a Google image search the other day for a specific car, I was offered a lilac-colored image that was not at all like the real thing, although it was rather intriguing and surprisingly useful to me; it will be in an upcoming post.
When I followed this image back to its source I was surprised to find that a company called Metro Molded Parts, which makes rubber and plastic parts for a huge array of older cars, is using an AI generator to provide images as well as written descriptions, performance, collectability appeal and other metrics for every car that they offer parts for, and that’s in the hundreds. It’s presented as a “Classic Car Guide” and of course threatens to pollute Google searches, which are already becoming ever more difficult in finding genuine sources rather than AI generated answers. I clicked on a sample of them and offer them here for your perusal, but you might want to try it for yourself.
Looks semi-convincing, at first glance. But a closer look shows that these images were AI generated too.
We’ll talk about the text information later, and look at a sampling of its images first. The surprising thing was just how few cars it actually nailed, or came close to that. Yes, it does know what the iconic ’55 Chevy looks like.
But it mangled the also-iconic ’59 Chevy by giving it an accurate ’58 front end. A nice hybrid of the two, but…
And yes, it’s got the ’65 Mustang down (almost) pat; the roof has too much curvature and there’s a few mother minor details, but pretty close.
And it’s got the 1972 Mustang down fairly well, except it thinks this is a 1975 Mustang II. That’s a pretty egregious mistake.
And the 2003 version looks like it got thrown in the washing machine on the hot water setting.
But its comfort zone is quite small, and it starts hallucinating all-too often; like much of the time.
A 1941 Ford Jeep (military MB)? Looks more like a Willys coupe to me.
But for the 1943 version, it’s trying a bit harder.
And for the 1945, it decided to make it a full convertible instead of a convertible sedan, although the roof is a bit low on both of them.
It seems utterly flummoxed by some very common cars.
And others it just mixes up. It’s got the shape of the ’68-70 Mopar B-Body down pretty pat.
Especially in the case of the Road Runner.
It just keeps using it too often, as in a number of years of Plymouth Furys, like this ’70.
Here’s another complete hallucination, with no similarity to the real thing.
It’s a fair bit closer with the 1960 Corvair. It would have made a sweet hardtop, as I’ve pointed out before.
But the gen2 is being seriously mixed up with the Camaro.
Hmm…where’s the wood?
The Edsel’s iconic horse-collar grille is totally missing.
The 1960 Falcon previews the ’63 Fairlane Sport Coupe’s roof.
This presumably is supposed to be the 1963 Falcon Sprint.
It does a lot better with the ’64 edition.
It should not be surprising that this is what it thinks a ’32 Ford looks like.
Or a 1926 Model T.
It’s better with the older 1916 T, although the body behind the front seat is pure imagination.
A “customized” ’58 Thunderbird.
The ’64 is one of the more relatively accurate ones.
But that’s short-lived.
It’s trying, sort of.
The real 1960 Valiant is so distinctive I assumed some that would come through. Not so.
Let’s try a few Lincolns. This 1957 has almost zero connection to the real thing.
The ’58 Lincoln looks like it’s got a bit of ’57-’58 Cadillac Eldorado Brougham in it.
And the 1960 is previewing the ’61, to some extent.
The ’61 is the closest one to reality. A coupe would have been a nice addition to the line.
By 1970 and in the next few years it tries to meld the Mark with the Continental sedan,
How about a ’57 Olds “888”?
Olds non-Achieva.
Henry J you say?
Looks like the Mirada got downsized into a little FWD sporty coupe.
Closer to reality again.
There’s vast numbers of trucks in this database. I picked a more obscure one, the big and tall Ford N-Series of mid-heavyweight trucks. I guess I shouldn’t be too surprised to see an F-100 for the image, as well as variations of that for quite few others from this period. I guess they haven’t scraped our article on the N-Series yet.
With every single car and truck, there’s accompanying data and this write up. A CC these are not. No, these N-Series trucks did not have Twin I Beam front suspension, which only came out in 1965 was only used on the F100 and F250 pickups. It’s bland, generalized, vague, misleading and sometimes just wrong.
But let’s not be too harsh on this poor overworked AI generator; it does know what a VW Beetle looked like, although this looks more like a ’62 than a ’46.
And the ’78 cabriolet has two tops, just in case!
Here’s the link again: https://metrommp.com/classic-car-guides-ratings-features/
Related CC reading:
A link to fuzzyman’s various AI generated “What If” posts
QOTD: Is AI Ready To Replace Human Writers At CC?
The Radioactive Cars of Chernobyl – Can You Identify Them?
What If: AI Generated Cars That Never Were by T.Halter
What If: More AI Generated Cars That Never Were (But Should Have Been) by T. Halter
I have to say that the 1969 Corvair/Camaro mashup looks pretty good.
“…a few mother minor details,” see even humans make mistakes.
Thanks for starting my week off with a laugh, Paul. Are the generated products of these AI gremlins mis- or disinformation? Misinformation, I hope – though those three-eyed ’60 and ’61 Lincolns look ominous. More radioactive cars, sort of like “Blinky” the three-eyed mutant fish from “The Simpsons”.
AI uses publicly available references to render images. If there are not enough references that it can find, then we end up with “hallucinations”. While I could see a lack of information for a more obscure model, such as the 1937 Cord, a 1965 Mustang should have lots of references producing a more accurate representation. This appears to be the case, but I would think that almost any vehicle after WW2 would have plenty of documentation. So the image of the hybrid 1958-59 Chevy is a mystery.
I see plenty of these hallucinations on social media; people who don’t know the difference are either creating or just distributing mislabeled (or just misinformational) pictures of non-existent cars. And when some of these posters are confronted, they get pissy for getting called out. I guess they don’t like being corrected/scolded. I think eventually, enough of this misinformation will become commonplace and younger enthusiasts will have no idea they’re being taken for fools.
Were I the owner of the website shown here, I would be mortally embarrassed. If the information put out by their media team is this bad, a potential customer could logically wonder what else is suspect about the operation?
As a user and maker of 3d models, I’d definitely reject this place. I expect most people who use these models in scenes are acquainted with the real thing, so the worst examples won’t sell.
You missed the wildest hallucination of all, the 1972 VW type 2! Air-cooled VWs that aren’t Beetles look too Beetle-like in general. It nailed the Rabbit though, with the sole exception of giving the 1975 model only a Simca 1100/Zastava 101-style slanted deck.
They caught this same website at The Autopian recently as well. https://www.theautopian.com/this-car-parts-site-is-a-dazzling-example-of-the-madness-of-ai-but-its-kinda-fun/comment-page-1/
Most of the ones I see on ‘Youtube’ are just fantasies. Lets get real. In the next year-if not sooner no American car manufacturer will make a vehicle that is NOT an SUV. These renderings should come with a warning-‘For entertainment purposes only’, so take it for what its worth.
That ’60 Dart has a Soviet vibe to me. AI may well be the death of us all.
Perfect source for images to be used in advertising where the vehicle cannot identified. I think Paul did a great CC on this. Most recent example is a Nissan TV commercial with where the sales prospect gets out of car no one would recognize – a Saab 9000 in this case.
We’re doomed .
-Nate