Home Creators Posts Import Register Favorites Logout

Content

Hi lovely patreons! If you've enjoyed a holiday over the Christmas period, I hope you had a fun/restful/exciting break with friends/family, or just spent a bit of quality time on your own, ready to tackle 2024! (I myself have spent a very enjoyable few days reorganising my workshop, which was lovely ;) )

I've also spent a bit of time playing around with generative A.I. I'm sure you're all aware of ChatGPT, Midjourney, Dall-E, etc. There's still some contentious issues surrounding the way in which their training data was obtained, the legal protection (or otherwise) of any content they create, and the ethical issues of automating processes that otherwise would have employed a scriptwriter, artist, or other creative person. But, there's no denying that the tools themselves are already very powerful, and will only continue to become more widely-adopted.

As you might expect, the area I'm particularly interested in is using A.I. to create "puzzle" art - particularly ambigrams - a sort of optical illusion that can be viewed in different ways to reveal a hidden message or subject, and I've been experimenting with different approaches which I thought I'd share with you. These images were all generated on my own PC using a combination of DeepFloyd and Stable Diffusion XL models:

"Squint" illusions

These illusions work by arranging features in a generated image according to a secondary control image. When viewed full-size, your brain pays attention to the detailed elements arranged in the generated scene. But, if you squint, stand far away, or look at a thumbnail or image on a small screen such as your mobile phone, the detail is lost and you instead focus on the more general arrangement based on the control image. 

Here's an image of a magnificent medieval walled city. But, if you squint, you should also be able to make out the image of Vermeer's "Girl with a Pearl Earring" painting.

The hidden image image can be anything you want - another image, a logo, text, or (for escape rooms...) a 4-digit code. Wonder what sort of organisation these people work for?

"Negative" illusions

These illusions work by simultaneous generation of a pair of black and white images of different subjects, but the "canvas" on which each image is generated is the negative of the other. On the left hand side there's some old photographs of people. But, look at the corresponding negative of that photo on the right and... it's a completely different person?! (yes - these image pairs are identical inverted copies of each other - you can take this image and invert the colours in any graphics program if you don't believe me!)

"Rotation" ambigrams

These are probably the most well-known style of ambigrams, where an image can resemble different subjects depending on whether it is rotated a different way up. In the left-hand column we have oil painting portraits of Victorian woman. But, turn the painting over and you see a depiction of a darker side to their character, shown in the right column.

The code and process I'm using to create to generate these images is still very rough and manual so it's not in any shareable state at the moment, but I'd be happy to answer any questions if anyone is interested in creating your own puzzle images like this, or perhaps want me to create one for you? 

And, as always, if there's anything else you'd like to see me cover - ideas for escape room puzzles or props - in any projects this year, please do let me know! x

Files

Comments

L Wo

Great post! I've considered putting AI art into my rooms, it's hard because when you're creating a room and want to put interesting things in it as visuals, ai tools can be a great way to keep it within budget. Even though 99% of your room is an original creation and represents months of your blood sweat and tears, certain people will only see the one photo in the room that was crested using potentially plagiarised work and have a tantrum, very frustrating! I would love to see you make a post about how you solve a lot of the issues that normally pop up when you create puzzles, like how you organise wiring or how you make it easier to fix things, what tools you use. Perhaps some interesting 3d prints that help you.

L Wo

Oh wow I've just seen those pictures, can you imagine a room full of those then you "do the scary thing in the room" and they all start rotating upside down. Totally changing the vibe of the room

Krista Jones

So cool! I especially love the squint puzzles. Do you know if they work as effectively when printed out/non-pixelated? How much control do you have over the foreground image? Could part of that be set, or purely from AI prompt? Thanks!!

Krazee Kraftz

Hi Alastair, I asked you a question about checking colors a while back and whilst you didn't respond personally to me, you did release a video that might have been your answer to my question not long after. Unfortunately, it wasn't helpful in my case. I tried to join the 3 groups on facebook you provided links to, waiting on approval at the moment, time zones! I do have another question about that prop but would rather not put the specific details in a general forum. So if you could possibly send me a message, I'll respond with my question, if you can come up with an answer I'm quite happy for you to share the generics of it without showing my prop directly. Jenny