Click to view attachment

I guess we should expect posts about AI derived cars as a semi-regular thing now.

Zach

I actually like it. It is a good progression of the 914 design.

Take the badge off the hood and it might not be too bad.

Proof AI doesn’t understand NARP.

I think they should have tried to echo the original fender lines from a 914 versus the big circle cut outs...

but not bad.

I like it but with the change @brant suggested. As soon as I saw the rendering I thought the exact same thing.

That badge is too big. Smooth design though.

Be grateful - if it was truly in the style of Dali, the wheels would be squares and probably melting off the edge of the road.

Just asked ChatGPT for an image of a 2024 Porsche 914 and this is what I got, raw!
Pretty sweet starting point for a auto designer to tweak.

Even if they built it, it would be half again bigger than a 914. Nobody builds small, light cars like a 914 anymore.

If you want an example of this, look at the original Mini vs the new Mini. The new one is quite a bit bigger.

Ain’t that the sad truth!

That looks pretty good in my estimation.

I have done design critique for my dad's stuff for most of my life.

What I see wrong in that picture is this:

As Brant mentioned the wheel arch openings suck. Front is not nearly as bad as the rear.

The proportions are off a tiny bit. The front is slightly too big from the windshield forward. maybe 5-10% too big. Car looks front heavy to me.

I think the targa bar should be more pronounced where it blends into the body at the bottom. Making this look more 911 like is a good call IMO.

Line at front upper door to fender is too curvy. Needs to be curved, just less.

No rear hood part line, windshield wipers, mirrors, etc.
It is an illustration...

I'd make the rocker panels more flared at the ends. Give the rocker a little bit of a coke bottle shape.

What's up with the rear spoiler/scoop looking this on the back?
I'd want to see more angles of that.
Looks funny at this perspective.

The fabricator in me sees no part line for front bumper/cap; needs one.

The technician in me sees almost no camber in the suspension alignment.

That would have taken my dad a week of 2-3 hours a night to produce.

His design would NOT have had that nice background and road.

I wonder how long it took an AI to do that?

Probably seconds...

The shape of the front end is excellent IMO.
We will be making a nose cap for the 916 clone and I will probably incorporate something similar.

So interesting to hear your comments! Yes, I would say the actual processing time to produce this was probably less than a second though I am not well versed in this particular technology. [just checked with Chat about this— you are correct “a matter of seconds”]

Here is the prompt ChatGPT created to send to DALL-E (guess I said 2023):

"A modern 2023 Porsche 914 concept car. The design should be a contemporary reinterpretation of the classic Porsche 914's design elements, featuring its distinctive targa top and mid-engine layout. The updated version should have a more aggressive stance with sleek, aerodynamic lines, LED lighting, and modern alloy wheels. The car should be placed in a setting that suggests performance and luxury, such as a winding coastal road or a sophisticated urban environment."

I asked Chat what would happen with minor tweak suggestions:

“If you request tweaks to the original prompt, the new image generated by DALL·E will likely have noticeable differences. The AI uses the text prompt to create an image based on patterns it learned during training, so even small changes to the prompt can result in different details in the image. The overall concept may remain recognizable, especially if the changes are minor, but the exact outcome can be unpredictable. Each generation is unique, and DALL·E may interpret the same prompt with slight variations each time.”

I then asked Chat to recreate the image incorporating the comments made in your critque:

Click to view attachment

Here’s the prompt it created:

"A refined 2023 Porsche 914 concept car addressing specific design critiques. Adjust the wheel arch openings to be less pronounced in the rear, and subtly alter the car's proportions to reduce the front section by 5-10%, balancing the vehicle's stance. The targa bar should be more distinct where it merges into the body, resembling the 911 style. Soften the curvature of the line from the upper door to the fender. Add necessary details like rear hood part lines, windshield wipers, and mirrors. Flare the rocker panels more at the ends, giving them a slight 'coke bottle' shape. Rework the rear spoiler to appear more integrated and less awkward from the current perspective. Include a part line for the front bumper to distinguish it, and adjust the suspension to show a more realistic camber."

So I think the lesson learned here is that this is not a viable design tool — at the moment. Nice idea generating tool but not tuned for tweaking.

So it spat out a modified Dino??

OK, this is the COOLEST thing I've ever come across in 50 years of being around an automotive shape designer (I cannot do this type of stuff myself nearly as well as my dad but he is a piss poor technician!) and to be able to produce "concept" type designs at a whim is awesome!

I liked the first one better, but it came up with yet another reasonable design in seconds...

Still look front heavy to me!

Thanks for sharing.

I cannot wait to show my dad this thread!!!!

It will piss him off, but he will instantly see the benefits to at least trying this...

I thought the Boxter was the evolution, or rebirth of the 914

...because it is. The Boxster/Cayman is the new 914 (but the 914 did both body styles at the same time)

I will let Chat chime in on this:

“The driving experience in the Boxster is more refined and accessible, but potentially less raw and connected than the experience of driving a 914. Enthusiasts often appreciate the direct and unassisted feedback from older cars like the 914, which can make them feel more engaged with the vehicle’s handling.”

I would love to hear his reaction!

Maybe you could have it design a "modern" 914 with an aircooled motor?

The huge opening in the front hints of some kind of water pumper motor & doesn't look right. Looks more like a Fiero GT (not that there's anything wrong with that).

DALL-E just doesn’t work like that at the moment. It would start again from scratch and one would probably not like something else in that iteration. It’s the nature of its input— textually description only, no starting images. And, further, you could give it the exact same text over and over and never repeat the design given.

Hmmm, last i used DALL-E, you could pick an image, erase a certain portion of it and ask it to redo that portion with something else, preserving the rest of the image.

Here you go, i uploaded an image i had created earlier (the one on the left), selected "Edit" from the menu, erased the face part and asked it to replace the face with an older looking one.

Click to view attachment

In the second rendering , the upper picture looks too much like a 911. The lower one is better, but the windshield opening is too rounded, and looks too much like a 356 roof line. It also needs a more pronounced targa band.

I would love to see the rear of these.

My only experience with it has been through ChatGPT-4.

Fiat 500

Dall-E….nah, how about a Dali car

Love this interpretation...



Click to view attachment

Oh wait. That's mine...

Wait . . . That’s a 914?

For the record,I desperately wanted a Chalon kit back in the late 80s.

No pop up headlights no care.

very very well done! Keeps the essence of the 914 design, while blending in nicely with the porsche lineup. I would purchase one!

I wonder if it would even be possible with today's impact requirements. Cars just seem to have to look 'bulbous' rather than sleek. The fiat 124 and Miata are as close as it comes to 'small and sleek' but they are still pretty bulbous. Even the new corvette has a big rear too.

yeeeep, as an automotive artist and designer, AI fills me with fear and hatred, haha. Even being particularly skilled in 3D and rendering, I can't create the CAD and output a render anywhere even close to what AI is running around with now.

On the flip side, I have much more control over the details.

Of the AI things I've seen, they all pretty much give a really nice and beautiful "big picture" but once you start looking at the details it all falls apart. AI is powerful, but I've sworn off using it, at least for now. I try to exercise my own creativity.

After seeing the two different illustrations it came up with it looks like AI is a better "back of napkin" sketch pad.

The number of quick napkin sketches I've seen from my dad is un-countable.

Everything STARTS from a napkin sketch...

Being able to refine the napkin sketch is where the art is!

This does not appear to be refinable.

It produces something new every time.

Still way cool, just not useable for extensive reworking.

Dad and I have been arguing about a retro-modern 916 nose cap for the front of the car.

His designs are too sharp-nosed IMO, and the one on the silver illustration is FAR more like what I wanted in the first place.

Imagine something like that but with a tiny bit more 916 in it.
Slightly more angle on the bottom of the side pods and a bit smaller grill maybe.

I'm fine with the "water pumper" look as the "916 clone" will be water cooled.

Thats what I'm after...

The first illustration really looks "Porsche" to me.

Back to in period “modern”. I think this interpretation is way cool.

Wonder which one caught your fancy Doctor…

Further attempts:
Click to view attachment Click to view attachmentClick to view attachment

I think the one that led this thread might still be the winner.

Looks like the limited prod Glockler 914

The Porsche Crest started on a napkin sketch as well; early '50s Max Hoffman & Ferry Porsche over a couple drinks in a Manhattan pub

Really nice illustrations!

The background would be harder to complete than the car.

OH, please, please ask it for an updated 916...

As requested:
Click to view attachmentClick to view attachment

that's 904...

Way cool.

Not 916. It doesn't know what a 916 looks like and it probably looked as close as it could and came up with 917...

Still look likes bad-assed 904s

Yes, requested 916 and got those.

I introduced my dad to the thread today.

He was blown away!

We will use AI in his designs from now on.

Not in the vehicles, in the back grounds.

The background work is as cool as the cars and FREAKING HARD TO DO by an artist.

His backgrounds will be AI now...

Thanks for sharing!

It was very interesting and entertaining.

The miss-designed 904s were very telling as well.

It would be nice to feed it an image to start from.

Ha! Simultaneously encouraging and damning. Thanks for sharing his response.

OMG! I love that 904 rendering!

Very nice renderings!

It's amazing what AI can do with these designs. But, why can't AI spell Porsche correctly? (check out the orange car above, and the car & building in post #38)

You're right!

It has misspelled Porsche a couple of different ways.
Wonder if it's a copyright thing. Probably is.

Use Ferrari in any published advertising without Ferrari's approval...

Just make a copy of an old Ferrari and they will visit, even if you call it something else.

I noticed the seats are not well rendered but look at the road surface in the orange 904 illustration. Freaking perfect! So hard to get both the shiny asphalt look and then the reflection of the car.
Very realist IMO.

If you are not familiar with the technology, it might be difficult to understand the errors that Dall-E produces.

Here’s what Chat has to say about it:

In DALL-E generated images, words or text elements may appear misspelled or jumbled for several reasons:

1. **Training Data Limitations**: DALL-E is trained on a vast dataset of images, but this dataset may not have sufficient examples of all words spelled correctly within images. As a result, the model might not learn the correct spelling for every word.

2. **Text Representation Challenges**: AI models like DALL-E are primarily designed for visual generation, not textual accuracy. The model might struggle with representing text accurately within an image, especially if the text is a secondary element in the image.

3. **Context Understanding**: DALL-E interprets text prompts and generates images based on them. However, its understanding of context and nuance, especially when it comes to text within images, might be imperfect. This can lead to misspellings or contextually incorrect text.

4. **Model Constraints**: The current state of AI technology, while advanced, still has limitations. Accurately generating text within images remains a challenging task due to the complexity and variability of written language.

5. **Image and Text Fusion Complexity**: Combining visual and textual elements seamlessly is a complex task. The model may excel at creating visual representations but falter when it comes to integrating text naturally and accurately.

6. **Randomness in Image Generation**: Part of the image generation process involves a degree of randomness. This can sometimes result in unexpected or inaccurate details, including text.

These reasons highlight the challenges faced in the intersection of visual and textual AI generation. As AI technology continues to evolve, improvements in these areas are likely to occur.