Before AI’s popularity, I never would’ve thought I’d be able to turn what’s in my head into actual art. We’ve truly been blessed with Midjourney, Stable Diffusion, and — of course, the model that started it all — DALL-E.
However, one thing I noticed about AI image generators is its inconsistency at generating real-life places. There are times where it would completely misunderstand the prompt. Sometimes, it would even hallucinate aspects of the landscape that aren't really there. Take this image from Midjourney using the prompt “The Great Wall of China” as an example:
It certainly could pass as A great wall in China, but it isn’t THE Great Wall of China.
Now, two of DALL-E 3’s biggest improvements are its nuance and accuracy. So, I had an idea…
Since I love all things retro, I thought it’d be a great idea to showcase DALL-E’s creativity by turning iconic landmarks into 8-bit pixel art. Not to get ahead of myself in this article but, let me tell you, they did not disappoint. And I’m so excited to share them with you.
From the gardens of Taj Mahal to the rugged walls of The Colosseum — here’s what the world looks like in 8-bit, courtesy of DALL-E 3:
Global Landmarks in 8-Bit
Statue of Liberty
This is an amazing reimagination of the Statue of Liberty. It’s bright and warm. All defining characteristics of her are present, from the bible to the color of the flame. If I were to nitpick a detail DALL-E got wrong, it’s that the pedestal should only have three windows instead of four. Other than that, I’d say that this is completely accurate.
Sydney Opera House
I’d say this is even better than the one before. Compared to the Statue of Liberty, this pixelated version of the Sydney Opera House has stronger lines and more supporting details. I particularly love how the sunset aesthetics and how well-defined the waves are.
DALL-E 3 really outdid itself on this one. This rendition of the Eiffel Tower is beautiful. The colors are vivid and punchy. It also showcases the generator’s imagination and nuance — I mean, I just can’t imagine any other AI image generator using the Eiffel Tower to separate the image into two parts so it could highlight the symmetry of the supporting details. Another detail I’d like to point out is how DALL-E used the empty space in the bottom to house the setting sun.
This is another accurate and gorgeous entry from DALL-E. It has earthy colors, a serene beauty, and it completely shows the intricate architecture of Taj Mahal. There’s also a slight tint on the art, helping establish the atmosphere. However, I couldn’t help but notice slight errors such as the guy sitting on the left missing his head or the mismatched reflection of the fountains and the Taj Mahal itself.
This is seriously impressive. The sun slowly creeping over the horizon gives character to the art as it paints the canyon with this warm, golden light. I also like that DALL-E also took some artistic liberties by setting the 8-bit art during sunrise. It helped accentuate the features of the Grand Canyon by providing contrast, allowing you to see the chasms in a different light.
In one word: Mesmerizing. A common thread among these images so far has been how good DALL-E is at generating supporting elements to the main subject. The snow-capped peak of Mount Fuji brings a wonderful contrast to the blue skies, the trees, the lake, and the pink hues of the Sakura blossoms. A small thing I love about this one is the ripples slightly distorting Mount Fuji’s reflection, showing both attention to detail and outside-the-box thinking.
La Sagrada Familia
I chose La Sagrada Familia as one of the prompts because of its intricacy and DALL-E did not fall short. It perfectly shows the quirkiness of the basilica’s architecture, from the unusual facade to its tree-like columns. You could also see the consistency of the shadows with the light coming from the left. However, I would’ve liked it more if the surrounding buildings were shorter, since the real cathedral towers over its neighbors.
Leaning Tower of Pisa
This version of the Leaning Tower of Pisa is more in line with old PS1 games rather than an intricate pixel art which, to be fair, follows the prompt better than the previous ones did. As for the image itself, I wasn’t completely blown away but it’s still pretty good and — most importantly — accurate.
DALL-E brought its A-game with this beautiful pixel art of the Colosseum. It’s like a scene from a classic film, only in 8-bit. My favorite part is how the light from the sunset hits the arches and casts shadows. DALL-E also avoided its tendency to repeat elements, instead focusing on creating accurate and uneven windows on the Colosseum’s fourth level.
Great Wall of China
DALL-E did a great job not just showing the distance and enormity of the Great Wall, but also its timeless aesthetics. I appreciate that it showed the weathered texture of the bricks which is crucial to the landmark’s unique and long history. I just wish it didn’t make the mistake in rendering the part before the first tower which resulted in two parallel flights of stairs.
I’m surprised at how vibrant DALL-E made this art of the Acropolis, considering that most images of it are muted and lack any defining details apart from the columns. But, it works. This is a unique take to the sculpture, showing even the nearby greenery and the tourists lining up to climb its hallowed steps.
Easter Island Moai
DALL-E really likes its sunsets and sunrises, and I don’t blame it one bit. The soft pastel colors help highlight Easter Island’s polynesian roots. The Moai are well-defined, but I would’ve liked more variety instead of copy-pasting the same face on each of them. I also wish that their sizes are closer to its real-life counterparts because (although I haven’t seen them yet) I highly doubt they’re twice the size of trees.
The Sphinx and The Pyramids
It got the nose right, that’s for sure. I’ll be honest, I’m not particularly impressed by this one, mostly because it didn’t show the pyramids more prominently. You could also see a rendering mistake on the Sphinx’s foot where there’s, well, two feet. The colors of the sky are great, but I wish they blended better.
Christ The Redeemer
If I were to pick a favorite, it’s definitely this one. I mean, would you look at that color? It provides amazing contrast to the hills and creates a beautiful tapestry in the sky. The blending of the sky and the clouds is a masterpiece in itself.
However, what sets this apart from the other artworks is the creative use of the sun by making it act as a sort of halo for the statue. No other AI image generator would’ve thought of that, hell, I don’t think that would cross the mind of most people. DALL-E’s nuance and creativity is really next level, and you can see it here.
That said, DALL-E did make a big mistake… It’s supposed to face the other way.
DALL-E ended strong with this amazing 8-bit pixel art of the Machu Picchu. This one took advantage of the pixelated medium to provide texture to the terraces. The way the morning sun gently touches the stone structures is beautiful. Even the mist-coated mountain top background is carefully crafted. I also like the addition of the alpacas, they’re pretty cute.
My Thoughts on DALL-E 3
I’ve used DALL-E 2 before and I gave it a mixed review because I thought it wasn’t creative nor smart enough to rival other AI image generators.
Now, it’s a testament to the OpenAI team that I’m considering using DALL-E 3 over Midjourney. In my honest opinion, I think it’s ten times better than its previous iteration and it really shows in these artworks.
It was able to turn iconic landmarks into something different, yet equally beautiful.
With generative AI only getting better, I’m really excited to see where DALL-E goes from here. Would it be able to surpass the popularity of Midjourney? Only time can tell.