I’ve tried using AI three times as a GM. Two were complete failures. The third worked well enough for my purposes in the end, but I certainly wouldn’t say it was good.
The first was trying to generate a picture of a very specific scene I had in mind. I wanted to show the players the landscape they were seeing. The AI was unable to generate a picture to my satisfaction, so I gave up. This was 3 years ago now so maybe image generation has gotten better, but it certainly wasn’t good enough at the time, and I haven’t bothered trying again.
The second was an attempt to generate some box text–style description of something. More out of curiosity than anything. It made some very basic factual errors. The description was of a statue of an archer, and it described her doing something with both her hands (hands on hips I think? I forget) as well as having her bow drawn. Humans don’t have 3 hands… I tried and tried repeatedly to get ChatGPT to fix its errors, but by the time I ran out of free credits it still couldn’t produce something that was even cogent, let alone artistically good.
The third is one that I think has the greatest potential. I had a great map that I wanted to use. I can’t remember if this particular one is one I paid for or if it was purely hobbyist put up for free. But it wasn’t printer friendly. I asked the AI to take the map and convert it into a printer-friendly format. So mostly blank white with black outlines of things, and a black grid. It really, really struggled with this task. After several back-and-forths I eventually got something that was…usable. But I was far from happy with it, and the AI regularly completely ignored the specific things I was asking of it to address the failure in the previous attempt. This was maybe 1 year ago or so. I think it has potential to get better because it’s not replacing a creative process, but a mechanical one. But even here it didn’t seem good enough yet.
I think youve pretty much nailed the ai experience in a nutshell - spend some time trying to coax a result out of the system over several iterations, finally giving up and reluctantly submitting half assed slop to a recipient that doesnt know any better.
For your pictures you could maybe try drawing it in paint and then get AI to IMG to IMG from that to add detail but generally following the hard edges fairly well.
Not done much with it beyond playing around to see what it can do a while ago. Civitait got blocked in the UK now and I can’t be bothered to open a VPN for something I didn’t really have that much use with beyond poke and see what happens.
![Be creative, don't pawn off the creative process [Rule]](https://slrpnk.net/pictrs/image/f9267fd8-c41a-4d21-8fc4-04a7456cbf4d.png){kind=link}
I’ve tried using AI three times as a GM. Two were complete failures. The third worked well enough for my purposes in the end, but I certainly wouldn’t say it was good.
The first was trying to generate a picture of a very specific scene I had in mind. I wanted to show the players the landscape they were seeing. The AI was unable to generate a picture to my satisfaction, so I gave up. This was 3 years ago now so maybe image generation has gotten better, but it certainly wasn’t good enough at the time, and I haven’t bothered trying again.
The second was an attempt to generate some box text–style description of something. More out of curiosity than anything. It made some very basic factual errors. The description was of a statue of an archer, and it described her doing something with both her hands (hands on hips I think? I forget) as well as having her bow drawn. Humans don’t have 3 hands… I tried and tried repeatedly to get ChatGPT to fix its errors, but by the time I ran out of free credits it still couldn’t produce something that was even cogent, let alone artistically good.
The third is one that I think has the greatest potential. I had a great map that I wanted to use. I can’t remember if this particular one is one I paid for or if it was purely hobbyist put up for free. But it wasn’t printer friendly. I asked the AI to take the map and convert it into a printer-friendly format. So mostly blank white with black outlines of things, and a black grid. It really, really struggled with this task. After several back-and-forths I eventually got something that was…usable. But I was far from happy with it, and the AI regularly completely ignored the specific things I was asking of it to address the failure in the previous attempt. This was maybe 1 year ago or so. I think it has potential to get better because it’s not replacing a creative process, but a mechanical one. But even here it didn’t seem good enough yet.
I think youve pretty much nailed the ai experience in a nutshell - spend some time trying to coax a result out of the system over several iterations, finally giving up and reluctantly submitting half assed slop to a recipient that doesnt know any better.
For your pictures you could maybe try drawing it in paint and then get AI to IMG to IMG from that to add detail but generally following the hard edges fairly well.
Not done much with it beyond playing around to see what it can do a while ago. Civitait got blocked in the UK now and I can’t be bothered to open a VPN for something I didn’t really have that much use with beyond poke and see what happens.