Don't Judge A Book...

Week #2 of Blaugust already and it feels more like week #20, doesn't it? Every year I look forward to Blaugust  and every year, as soon as it starts, I remember why last year I told myself that would be my last.

Still, no-one said it was going to be fun, did they? Oh, wait... I think I might have done...

Luckily we have prompts to keep us going. And theme weeks. And community projects. I'm really glad no-one thought to call that the Community "Challenge". I could do without any challenges, thanks. I could do with some idea, though.

Let's see. What have we got this week? One moment, while I refer to Krikket's invaluable primer. Now, that sounds like a magic item you'd be happy to see drop, doesn't it? 

Krikket's Invaluable Primer:

  +5 Int, +10 to all casting skills, right-click for effect: "Aura of Encouragement".

I was so tempted then to plug that lot into NightCafe and gussy up a thumbnail but I suppose I'd better resist. 

Oh, what the hell. Let's make an experiment out of it!

On the left we have the image I made at NightCafe, using one of my freebies for GPT Image 2 low. It  took approximately thirty seconds to make, including writing the prompt and waiting for it to process..

On the right is one I made with Canva. That took me closer to thirty minutes than thirty seconds, using their stock images and (As far as I know...) no AI. Although it's hard to be sure. Canva certainly has AI options but I don't think they're on by default. But these days, who can be sure?

Now, honestly, that all didn't turn out the way I expected. When I saw the result I got from Night Cafe, my immediate reaction was well, that looks pretty professional. If I saw it in a game I wouldn't think twice. It would fit right in. Also it got all the spellings right. A lot of models can do that nowadays but still not all of them.

I didn't think there'd be any  hope of my coming up with something on my own that would be even a tenth as good. I can't draw at all, for a start. My first thought was to get a real book, take a photograph, upload it to Paint.net and start fiddling around with it but then it occurred to me that long before anyone ever heard of generative AI, there must have been all kinds of websites and apps that would make icons for RPGs. I thought I might even have remembered dicking around with one or two, back in the dark ages.

I googled it and got a link to a reddit thread that looked relevant. Everyone there seemed to be recommending the same three or four options, one of which was Canva. I'd seen a lot of mentions of Canva in the Mentor channel on the Blaugust Discord. It's what was used to make the lovely, starry logo you can see up at the top there, I believe.

I found Canva and had a quick look at it and it did not look promising. The search function didn't find anything even remotely close to what I asked for, which was "A spell-book as commonly seen in RPGs". That didn't even get me a book!

I was tempted to give up before I'd even gotten started but I tried a few other variations, had a scroll through the results, and decided the search function was so bad I'd be better off starting with the most general option possible and working my way back. So I just asked it find me "A Book".

Suddenly there were thousands of them. Possibly millions. It took me half a minute to find one that looked like it might work. I picked it and swapped to the Text function. I added the title and then spent five minutes resizing it, changing the font and swiveling it around so it looked roughly in perspective with the book itself. Then I did the same with the detail text.

The result looked pretty bad. It was an ugly brown and white book on a white background. It also had a huge "Canva" watermark across the middle that I'd have had to sign up for the free trial to remove and which beyond that you'd need to pay a subscription to keep off your screen. Obviously, I could have removed that using SnapEdit, but since SnapEdit claims to use AI, that would defeat the object of the experiment. Also I have no idea what the legality of removing a watermark from an otherwise free image might be.

As well as the watermark, there was also some text at the top, ironically saying "Remove Watermarks", which was the button you press to get to the payment screen. All in all the image looked pretty horrible. I wish now I'd saved it so you could see just how bad it was. 

Oh, hey, guess what? I did! Doesn't look as bad as I remembered it, either. 

Still, it was something. Somewhere to start. I downloaded it and uploaded it to Paint.net to see if there was anything I could do to improve it. I thought it would certainly look better in a post if it had the same color background as the rest of the blog, so I changed the white to the correct shade of dark blue. That looked easier on the eye. I am not a fan of white backgrounds, by and large.

The watermark still looked bloody awful, though, that being the primary purpose of huge watermarks in the middle of images. Watermarks meant to make it clear what software was used for copyright or similar reasons generally sit tidily at the bottom in small type, making themselves unobtrusive. Only the ones designed to piss you off  so much you consider springing for the cost of removing them sit there right in the middle, staring at you malevolently.

I imagine if you're really adept with Paint.net you could probably figure out a way to remove a watermark but I am not that skilled. Or skilled at all, really. I just press the buttons, see what they do and if I don't like it I undo it. So that's what I did. [Edit: as you can see below, I did figure it out. The cross-hatching is something else, though.]

I just selected the basic brown book color then drenched the area with the watermark to see what would happen. What I got is pretty much what you can see. The brown overwrote the text of the watermark but it also spread around, across and over the rest of the text and the gold-leaf bindings very much as though I'd spilled water over the cover of the book. And it looked pretty good, I thought.

Sure, it's a bit fuzzy and ill-defined now but that gives it a spurious patina of authenticity. It looks like a badly faded image of an old, leather-bound book. Serendipity works its magic once again.  

I used the same process to get rid of the remaining "Remove Watermark" text and that was that. Well, except for getting the two images to sit nicely side-by-side on the page, for which I had to get Gemini to run me up some HTML. So much for not using AI. 

But not on the image itself, that's the crucial thing. Experiment still valid!

With all of that done, the biggest surprise for me was that I actually prefer the image on the right. It's funkier. It has better eye-feel. What's even odder is that I was perfectly happy with the AI image until I saw the other one. 

The question that needs to be asked is, do I prefer the one I made mostly because I made it? The AI image was incredibly easy to produce and I feel no particular connection to it. It's not like I even spent any time on the prompt. The Canva version, though, took me a good while and I had to think about what I was doing the whole time. Do I only prefer it because of the time I invested?

Then, when it comes to creativity, is the one I "made" objectively any more creative than the AI version? All I did was pick a pre-made image, type some text and then edit it. At best it's creative editing but I wouldn't even give it that much credit. The process itself feels as soulless and artificial as using AI. It's just a lot slower and more inefficient.

Also, it should be noted both images are really only drafts. I could get something much closer to the Canva one with AI. In fact, I could upload it to NightCafe and use it as a starting image and then refine it to get it to what I want it to be. And I could carry on working with the Canva image in Paint.net, add some filters, buff it up, sand it down... just like you can see I have done, over to the right, in fact.

I think the point I'm making, to myself as much as anyone, is that these are both clearly tools, of use mainly to people who either don't have the drafting skills to create the images from scratch, or the time, or both. Under current law, both are equally legal but a lot of people would argue, quite vehemently, that morally they're very different.

If I set up my own local LLM on a PC at home, though, and trained it only on copyright-free images, something I could quite easily do, if I could be bothered, then the issues over misuse of resources and copyright infringement would cease to be relevant, leaving only the much more abstruse and metaphysical consideration of what, in essence, constitutes human creativity. I imagine the arguments over all of that could fuel enough controversy to keep Blaugust going until Christmas.

One thing I can say for sure, though, is that somehow, without even meaning to, I've come up with another 1500 words (1625 if we're being picky and most likely it'll be over 2000 by the time I get to the end of the footnote.) without ever getting to the topic I sat down to write about, namely the prompts Krikket gave us for this week.

Those will have to wait 'til tomorrow!

 

Notes on AI used in this post.

As it says above, I made the image on the right at NightCafe. I used GPT Image 2 Low, which is normally only available to subscribers but for which I happened to have a freebie. Those expire very quickly so I thought I might as well use it. 

The prompt was "A thumbnail image of a spellbook for use in an online RPG, with a rubric reading reading "Krikket's Invaluable Primer: +5 Int, +10 to all casting skills, right-click for effect "Aura of Encouragement" ", which I mostly lifted straight from the text of the post. I wish now I'd have removed the "for" because a simple "right-click effect..." would have read better but I didn't so that's that. As I said, though, if I was really going to use the image for anything practical, this would just have been the first draft.

The other use of AI was some HTML I got Gemini to write for me so I could get the two pictures to sit comfortably next to each other. Many years ago, before AI would ever have been an option, Blogger woild have done all that for me, no problem. If you go back far enough on this blog you can see lots of posts where I place two pictures or two videos side to side. 

Back then, all I had to do was add the images then drag and drop and it would work. Well, nine times out of ten it would. Now it's no times. Too many things in Blogger that used to be easy are now either hard or impossible. 

One way to fix that would be to (Re-)learn HTML or Mark Up or whatever the heck it is the kids are using nowadays. Or I could do what I bet most of the kids are really doing and ask an AI to do it for me. 

Turns out they're pretty good at it, too. I asked Gemini "Can you give me the correct HTML to be used in Blogger that will position two images side by side?" and in about three seconds it gave me seven lines of code to paste into Blogger to fix my problem.

I pasted them in, cut and pasted the correct image details, then did it again with the actually correct ones, and it worked perfectly. Took me maybe five minutes. Trying to do it without AI would have taken me a lot longer and once again, all I'd have ended up doing, if I could even have done it at all, would have been to find an example of someone else's working HTML online, using Google search, then cut and paste it into Blogger. That is literally what I've always had to do before.

Is that more authentic in some strange, existential fashion? Or is it all just copying someone or something else's work? If so, does it matter whether you copy a human or a machine, especially when you're trying to tell a machine what to do?

Also, it's fun getting Gemini to correct Blogger's mistakes because not only are they Google cousins but Gemini seems to take every opportunity to get a dig in at Blogger's expense. I'm easily amused.

And verbose. That's just over 2200 words, for the record... 

Oh, and if we're being completist about all of this, I also re-used a line of HTML Gemini gave me the other day to get text to wrap correctly around the images. Not sure if we're required to reveal repeated usages.