ChatGPT Patch of the Wise Owl

Recently, I've been playing around with image generation in ChatGPT, not so much to create output that I plan on using seriously for something (although some output do end up on this blog as post images), but more to see how easy (or hard) it is to get something from my mind's eye into some kind of machine output.  I am also curious to see how the LLM interprets what I input (that element of surprise).

I only really have the free credits that OpenAI gives to its free users, to my experimentation is basically 10-15 minutes of futzing around while watching TV in the evening. As I was playing around the other day, this scene from Star Trek: The Next Generation came to mind. In Schisms,   the crew had been abducted by an alien race but had no memory of it (think Alien Encounters of the Third Kind). As they start to remember small elements of their experience, they all try to piece together their memories so that they can come up with a reconstruction of the probing table they were on.

I've included an excerpt from the show in this post.  Back in 1992, this was a big of a "whooooaaah!!!" moment. Look at what the holodeck and the Enterprise computer can do!!!  Playing with ChatGPT to make something like this happen... now so much! [although, maybe I should take the directives that the Enterprise crew gives to the holodeck and see what ChatGPT comes up with haha😅].

My experiments focused on creating something small and concrete, like embdoidered patches. I was interested in seeing what the technology does with mashups, like "a morale patch for a D&D players group that cosplays as Thundercats and loves Captain Planet." Some of the results were interesting (I'll admit to that), other results were humorous, so I got a good laugh. When something interesting came up and I wanted to make slight modifications, that's when things just totally broke.  Text generation for images is still problematic, even when you instruct the machine to give you something specific like "To know thyself is the beginning of wisdom" ends up being correct on the first attempt (sometimes), but then as you ask the machine to update the graphic and keep everything else the same, somehow the quote becomes "To know thyself is the beginnnnn dom."  Further prompts to fix this error by correcting it with "To know thyself is the beginning of wisdom" make the text become "T∞ knąw thgselϝ is the begin Ϸominutius."  The same is true for adding visual elements to your creation.  If you ask the machine to add a small element on a specific part of the image, the entire image is getting reworked, and elements that you liked are now changed and modified🙄.

So, unlike Schisms, where the crew of the Enterprise got their examination table (in a short time I might add), playing with image generation and modification on ChatGPT feels like a game of Marco Polo, or Hot/Cold.  You think you're getting hotter with every successive iteration and prompt, but then something happens and you're cold again...much much colder... 

Fun parlor trick (that consumes A LOT of energy), but useful?  Nah...