Welcome to the 2nd in a series of posts where I will test AI image generators and see how they handle making maze art. I will be asking 10 prompts and seeing what gets generated. My goal is to evaluate different AI image sites against each other to see how they perform. In my first post I discussed the project, and today we start with the first site evaluation: DallE2.
Update from 2026:
When I did this series in Spring of 2023 I found that none of the text to image generators were able to generate actual mazes, including this one. They were able to make images that looked maze-like, but none were solvable puzzles. The 10 prompts below are a record of what was possible then, but each generation of AI pushes the limit further on what is possible. Feel free to try some of the prompts below to compare how far the development has come. All of these were made using text to image generation and not coding.
Making Maze Art with DallE2 / CHATGPT
2023: You can access the Dalle2 website here. You must sign in. As of this post there is no limit to the number of prompts you can ask for, and the site is free to use. Each prompt, or generation with DallE2 will generate 4 image options. For this exercise I chose the one closest to what I had asked for, or the most interesting. Let’s get started.
2026: I used ChatGPT to generate comparisons for the first 3 prompts below. I had to sign-in. Dalle text to image is imbedded in the product. I used the free version. Gallery images are labeled below. Use the next/previous controls to compare 2023 vs. 2026 images.
Prompt 1 - Make a medium difficulty maze of the Eiffel Tower in black and white with arrows at the start and finish
2023: I see the Eiffel Tower. In fact I see 2 of it for some reason. I see a maze like structure. It is black and white. But I see no arrows, and the maze is not really a maze. There are random maze lines, and marks around the page that distract from the image.
2026: I see the Eiffel Tower that is much cleaner in shape. I see a maze look inside the structure, but the maze is not solvable. It is in black and white. There are arrows this time but they are non-sensical.
Verdict: The generation improved are moved closer to what was asked for, but an actual solvable maze remains elusive.
Prompt 2 - Draw a medium difficulty large maze of the Empire State Building with the start and goal embedded in the structure
2023: Not sure what this is. I changed the prompt vs. #1 to get rid of the arrows…I like the mazelike structure on the top of the page. And there looks to be a small Death Star under it. I think maybe DallE2 can’t make mazes.. just mazelike drawings ?
2026: So much better. We see the Empire State Building. We have a mazelike internal, but still not solvable. We get large Start and Goal arrows which do not make sense, especially the Goal, but are almost there.
Verdict: The 2023 generation was so bad that anything would be an improvement. 2026 was much better and gives me hope that this will be possible soon.
Human Made Maze: I made a huge Empire State Building maze around 2015 using Microsoft Paint.
Prompt 3 - Draw a difficult maze of the White House pixel art style
2023: It is pixel art style. Despite having no start and goal it looks the most mazelike of anything generated yet. And there is a house, although it is grey and looks like a dirty igloo. Interesting. Not winning any awards.
2026: This looks like an image of the White House in pixel style. Th internal section is too open to be considered a maze. And the start and goal arrows are awkwardly placed - the AI does not understand how they work.
Verdict: Really great pixel art of the White House in 2026. Much improved, but still no solvable maze.
2026 Human Made Maze: I made a maze of the White House (not pixel style) that you can see on it’s maze page and download it for free. I have considered doing a human vs. AI challenge. I would say I won this round.
All remaining prompts come from 2023. A summary from 2023/2026 is available below.
Prompt 4 - Draw a difficult maze that looks like a drawing of a famous building in sketch style
After asking for 3 famous buildings and seeing DallE2 go 1 for 3 I let it pick it’s own famous building and decided to try ‘sketch’ as the style. It got the style down, it looks mazelike (but not a real maze) on top of a large building with an arch entryway.
Prompt 5 - Draw a maze in the style of doyoumaze.com of a skyscraper in NYC
Nope. Maybe I’m not well known enough.
Prompt 6 - Draw a maze in the style of Sean C Jackson of a scene from a large outdoor market
So we try a famous artist of mazes. I don’t see an outdoor market but I see a lot of interesting squiggles. The maze-like shape from this perspective looks cool. Maybe there is an outdoor market there with some imagination.
Prompt 7 - Make a maze of a slice of an orange in color
I simplified the ask and got the best technical answer yet. A start and goal would even make this a real maze !
Prompt 8 - Make a maze integrated on top of a photograph of a king sitting on his throne looking cantankerous beside his beautiful queen
Cool 3d looking maze in the background ! The king is beside his queen…but really behind his queen. The look on his face is something more Frankenstein than anything else. And the hands are a problem (of course).
Prompt 9 - Make a solvable maze that is very large and very difficult to solve because it is so complex
I wanted to test the maze only capabilities. Fail. What is with the red and grey items throughout ?
Prompt 10 - Make a 3d render of a red and blue glossy cube maze
4 shown because all were interesting and none was better than another. I feel like these images would be fun to walk into.
Summary and Conclusion:
2023: How did Dalle2 do ? Eh. It can’t make mazes, just maze looking structures. So you get some interesting outputs, sometimes, but nothing great. The 3D portion is worth exploring more !
2026: The image generation is cleaned-up with no random lines or images imbedded. Every building asked for looks like the building. Mazes look clean, but are not actually solvable puzzles. The generator has trouble understanding what start and goal arrows are for. Humans are still ahead when it comes to making maze art.
Coming next: Making mazes with AI: Stable Diffusion
