• Member Since 25th Feb, 2013
  • offline last seen January 27th

Titanium Dragon

TD writes and reviews pony fanfiction, and has a serious RariJack addiction. Send help and/or ponies.

More Blog Posts592


This is not a Gryphon · 7:36am Aug 29th, 2022

Some years ago, a famous French painter made a painting of a pipe, labelled "Ceci n'est pas une pipe." - "This is not a pipe."

And he was telling the truth. As he pointed out, it was a painting of a pipe - you could not put tobacco in it and smoke it, so if he had said it was a pipe, he would have been lying!

Called "The Treachery of Images", this is a famous work of art exploring the false nature of art.

So, in that light - this is not a painting of a gryphon on a piece of paper.

Nor is it a photograph of a painting of a gryphon on a piece of paper.

Nor is it even a digital copy of a photograph of a painting of a gryphon on a piece of paper.

No, this is a piece of AI generated art created using MidJourney's latest test, using the prompt "Handsome regal gryphon by Jordan Nelson" (and a lot of work figuring out what to tell it, and "rerolls" - the AI does not generate pieces this nice every time).

And no, it's not a photograph of not a print out of a piece of AI generated art; the entire thing was generated, including the "photograph" of the piece.

I've been having fun with MidJourney. It's a neat toy, and I've made on the order of 7,000 pieces with it. Most end up not so great (it takes a lot of trial and error and reworking to get it to actually generate what you want), but some are pretty cool, and I've been using AI generated art in my D&D games and posting them to FA.

If you're curious about what I've been up to, my gallery is on MidJourney's website, and I'm working on getting some pieces up elsewhere.

I hope you all had fun at Everfree Northwest this weekend!

Comments ( 10 )

Nice gryphon

Damn, those are some nice lookin' pictures that AI made. :raritystarry:

I loved the lamb one

Getting this sort of program to produce what you want is like knowing how tweak Google search requests, but even more persnickety. Great work in tricking the electrons into doing what you want. The coin where it looks like the griffon is lunging out of is my favorite, but the diagrams are a very close second.

Ok Love the rooster on the coins. That program sounds fun.

I can't be the only one who imagined somebody rolling a blunt out of The Treachery of Images, right?

Yeah, basically. The AI isn't actually intelligent, so it's all about trying to figure out how to phrase what you want correctly. But because there's also an element of randomness to it, you sometimes don't even know if your prompt was "right" and you just got unlucky, or if you are going down the wrong track, so it involves a lot of experimentation and trial and error and refinement.

It also is better at generating some types of images than others, and figuring out which prompts more consistently yield good results is part of it as well; when you get the initial output, it generates a 2x2 grid of small thumbnail images which you can then upscale into larger ones (most of the ones above are small compared to the actual output of the program, though it isn't the best at detail work).

Some things are impossible or near impossible for the program to output, and you can only get so specific because it doesn't really understand English so prompts can easily "confuse" it. It also has trouble segregating objects, so if you ask for a, say, hound wizard teaching their fox apprentice how to cast magic, there's a good chance both the figures will end up foxes or hounds, or some amalagation of the two, or you end up with three figures, etc.

Another example: I can ask the new test version to make gryphons, but if I ask it to generate gryphons with spots like a snow leopard or feathers colored like a blue jay, it will just spit out a snow leopard or blue jay at me.

As such, it is far more of a toy than it is a practical art program in a "replaces commissioned art" sense, but it can generate a lot of more "generic" art - I use it a lot for generating art for NPCs and dungeons in D&D and it mostly works pretty well for that, though the latter is basically stitched together (like asking it to generate some floor tiles, then asking it to generate doors, etc. then using the magic of photoshop to stich them together into a battle map to play on).

It takes both skill at writing prompts and patience to experiment and engage in trial and error to use, and how much depends on what particular thing you are trying to accomplish with it. There's also multiple "models" of it right now, and they are good at different things; I created the gryphon and the coin using a different model than the other pieces above, as one model is better at generating fake photographs and photorealistic objects while the other is much better at making more artistic things and things like anthropomorphic animals.

The codex that you liked is something I figured out to make a while ago, and you can actually make fake "books" which contain nonsense AI "text" about whatever subject matter you want. I've created ones "about" dragons, aliens, magic items, summoning circles, greek ruins, egyptian ruins, etc. Once I figured out the "trick" to making them it was easy to manipulate it into "theming" them how I wanted.

Thank you! I'm glad you liked it!

She is a cutie, isn't she?

Thanks! Yeah, it is fun, and I enjoy playing around with it. It's basically a toy that makes cool things.

I suspect that would not make for a very good joint. :ajsmug: Not sure if they used lead based paint, either. :twilightoops:

Or is like you play ai dungeon

Will not lie, Magritte is one of my favorite artists. Ever!

Login or register to comment