Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[neobit]

This is sick already.
I tried several of such tools in past years, but what Midjourney is currently offering is stunning. The revolution to art is on the verge and is both frightening and exciting.

Before I go too far, just look at this example.
I used this prompt twice to see what it will create:
a dwarf wearing cave spider silk coat and holding too heavy pickaxe at hand, he has a long white beard and looks distressed, blurred background
and got those two series of images



Of course, it is not perfect yet. It is slow, not precise and sometimes basically wrong, but what we see here is more than what I would expect.

Imagine a tool that would be trying to:
- use a life of a character from legends and create images for every entry of their life. It would not be precise but for sure will give more for those that care about graphics and more depth like that,
- images of artefacts when they are created,
- forgotten beast coming not only with their description but also graphical interpretation....
and much more...

Then you would be able to parametrize it to make the style pleasant for you... so not only it will be creating all those images, they will be unique for you also by style.

I work in industry for more than 20 years and still it is overwhelming for me. I do not see it as a replacement for art, but as an extension.
We will become AI operators and expand our perception and creativity to another level.

p.s. I know that for many this is a wrong way to go as some sees it as a killer of creativity (similar to what movies are to books). I certainly hope it is the contrary.

Cheers!

[Skynet]

I work in industry for more than 20 years and still it is overwhelming for me. I do not see it as a replacement for art, but as an extension.
We will become AI operators and expand our perception and creativity to another level.

Not a replacement for art, but maybe a replacement for artists. After all, if humans are "AI operators", then that means the AI is the one making the art - all the humans are doing is giving editorial guidance and curating outputs, but that's probably less prestigious compared to the actual creative labor done by the computer.

But yeah, Midjourney is super-cool. I should try to use it, and combine its output with AI text generation.

[neobit]

I don't think it replaces artists. IMO it enables art in a one way or another for a wider group of people.
As for existing artists, it will change the paradigm, but eventually artists will just use it as an extension to existing tools (e.g. base for art or a way to get ideas). Like digital cameras to analogue ones had so many things automated, people were claiming it is not art any more. Any smartphone now has a master camera inside. I have seen so crazy things in programs for making music e.g., automatic drums, autochords progression, same in graphics or photography like auto replacement for the skies in background and more. Those were just tiny steps towards what we see now. And this is just a beginning…

I agree with one thing though, people will need to adapt to such things.
Also, I might be too optimistic and simplify the dangers here as I work in automation for so long now. We have seen revolutions in industry several times... not sure a single one was such a big step as AI can be.

p.s. For now though, I am hyped :=


p.p.s. your nickname just puts a dark clouds over this topic

[Cathar]

Not a replacement for art, but maybe a replacement for artists.

It depends what you are using the art for, really. For single shot illustration meant to be seen out of focus from afar ? Maybe. For a long term project good luck having your AI draw the adventures of a specific dwarf with a specific apparence.
And as a store of value/ speculative good AI will never replace artists. The worst, untalented new-age abstract non-figurative artist with a bit of networking will always produce better stores of value than something that works automatically. The world of art is heartbreakingly cynical when it comes to economics, but even them understand that if people start giving NFTs to museums to deduct their value from taxes, the government will take action and nobody will like it.

I now use AI in my art, actually. Works great, makes texture and some details that are a bit bothersome to draw/paint manually. It's good for what it is - I sometimes need to generate wonky designs or scribbles that look like text while being unreadable. Logos also work great, and it's also a good starting point to get idea. But people believing AI will allow them to bypass the learning process in art are up for a rude awakening I fear.

[neobit]

(...)For single shot illustration meant to be seen out of focus from afar ? Maybe. For a long term project good luck having your AI draw the adventures of a specific dwarf with a specific apparence. (...)

That is precisely the problem we had even with Dungeon AI. The inputs are generally volatile and really dependent on the current state of the learning machine. I think we won't be able to quickly overcome this. Typically, there are ways to use other images as the input to the generation of new one. However, until the system can correctly distinguish what is what on the input image it will be treated rather as a theme, not a source of information of real objects.

That being said, I wanted to put my theory to work.
Legends are mostly unusable in current state. They contain too much information on names and too few about the situation/action of the event. I don't see a way to do it correctly, even with curation of those entries.
Artefacts worked a little better. Description, as is, cannot probably be used directly as it also contains a lot of names etc. After curation and some work with priorities on the parts of the description, results were sometimes good. What I liked is that they give at least some 'taste' of the thing.
Let me present to you some examples. Curated text (removed names, simplified description) with priorities (giving parts of the description bigger or smaller importance).

bronze spear, it is encrusted with oval tube agate cabochons, it is adorned with hanging rings of orpiment, on the item is an image of an onion in kaolinite

The top-left is a complete miss, but others are pretty cool as they are.

bismuth bronze crossbow, encircled with bands of trifle pewter

Just horrible. One can probably believe the items don't necessary be similar to what we have in our world, but in this case acceptance must be big.
I tried to rerun it again then with same parameters.

A little better, but still pretty funky.

schist bracelet, encircled with bands of oval schist cabochons, on the item is an image of square cut gems in schist, on the item is an image of an octagon cut gem in schist

This was rather easy one, and I would say it is satisfying.

statue of a dwarf and dwarves in gold::2 dwarf is surrounded by the dwarves::1.5 artwork relates to the anscension of the dwarf necromancer to position of queen in fortress::1 encircled with bands of gold::0.7 object menaces with spikes of tetrahedrite and red zircon::07

This one uses priorities. The effect is far from what description tries to depict, but as an artistic interpretation I like it.

silver battle axe::2 studded with silver and encircled with bands of round diorite cabochons::1 adorned with hanging rings of schist::0.5

Bizzare again.

Recreation made it a little better.

almond wood cup::2 studded with copper::0.2 decorated with almond wood and encircled with bands of platinum and oval chalk cabochons::0.1 object menaces with spikes of chalk and octopus leather::0.1 on the item is an image of the goblin striking giant leopard in bismuth::0.4

Fine. Nothing of the image, but probably I could use the better priority for it.
almond wood cup with image of the goblin striking giant leopard in bismuth::2 studded with copper, decorated with almond wood and encircled with bands of platinum and oval chalk cabochons, object menaces with spikes of chalk and octopus leather
So I changed it a bit, changing the original structure and giving the cup with image bigger priority than details.
The effect was interesting, even if still not precise.


almond wood blowgun::2 decorated with bayberry wood and encircled with bands of elk bone::0.2 menaces with spikes of cave spider silk::0.1 on the item is an image of a dwarf::0.4

Very cool.

We can say, that artefacts works rather fine. As mentioned before, not precise, sometimes completely wrong but for people like me that care so much on added visuals they give more depth. One can argue whether they add spice to them, as those items not necessary are similar to their counterparts in our world.

Lastly, I thought I will check some dwarfs description (also curated). Results are probably the worst, probably often they would mislead as they seem to be wrong.
Let's see.

male dwarf::2 lonely::0.2 he is thin and scrawny::0.5 his sideburns are clean-shaven::0.5 his very long moustache is neatly combed::0.6 his very long beard is braided::0.7 his very long hair is neatly combed::0.4 his ears have large hanging lobes::0.4 his eyes are turquoise::0.4 his nose is short::0.4 his hair is sandy taupe::0.4 his skin is pink::0.4 his lower body bears a tiny straight scar::0.4


female dwarf::2 lonely::0.2 her ears have great swinging lobes::0.4 her hair is clean-shaven::0.4 her eyes are turquoise::0.4 her pink skin is slightly wrinkled::0.4 her hair is dark::0.4


female dwarf::2 lonely::0.2 her ears have great swinging lobes::0.4 her very long beard is braided::0.7 her hair is clean-shaven::0.4 her eyes are turquoise::0.4 her pink skin is slightly wrinkled::0.4 her hair is dark::0.4


female dwarf::2 lonely::0.2 she has incredible muscles::0.4 her nose is extremely narrow::0.4 her very long beard is braided::0.7 her medium-length hair is neatly combed::0.4 her eyes are red::0.4 her pink skin is slightly wrinkled::0.4 her hair is gray with some white::0.4


male dwarf, he is thin and scrawny, his sideburns are clean-shaven, his very long moustache is neatly combed, his very long beard is braided, his very long hair is neatly combed, his ears have large hanging lobes, his eyes are turquoise, his nose is short, his hair is sandy taupe, his skin is pink, his lower body bears a tiny straight scar


female dwarf, her ears have great swinging lobes, her hair is clean-shaven, her eyes are turquoise, her pink skin is slightly wrinkled, her hair is dark


female dwarf, her ears have great swinging lobes, her very long beard is braided, her hair is clean-shaven, her eyes are turquoise, her pink skin is slightly wrinkled, her hair is dark


female dwarf, she has incredible muscles, her nose is extremely narrow, her very long beard is braided, her medium-length hair is neatly combed, her eyes are red, her pink skin is slightly wrinkled, her hair is gray with some white


You would need to go through those to see many flaws with images. It is OK, but not very satisfying.

Summary
What would be cool to see are those with usage of the style unique for the game—it is possible to stylize it already, but I didn't want to impact too much on the result yet.
Probably would require better prioritization of parts of the description, maybe also better curation of it.
And definitely a script/mod that can easily export and maybe even curate the text from the game to the input of the machine.

Sorry for the long post, but I really wanted to share with you of the capabilities of the tool.
I absolutely will try to use it for my next play through of Dwarf Fortess - manually though.

p.s.
look at some not related to DF pictures I have done today that are fascinating - I leave links instead of embedding images as they are huge
https://mj-gallery.com/e1bf4de6-443a-47a6-ad8e-8f19e3f89fa8/grid_0.png
https://mj-gallery.com/280bb842-aeff-4d53-88cc-e10bc397880e/grid_0.png
https://mj-gallery.com/f276357a-55fe-4d37-ba62-f70b7a5ae721/grid_0.png
https://mj-gallery.com/f0b6e072-601e-4ff0-99e1-b6509e79cda8/grid_0.png
https://mj-gallery.com/e8d00109-cfb7-4a2e-a914-dad882b7ae50/grid_0.png
https://mj-gallery.com/4bdaa123-8dae-4d28-9c2c-7d3f4711222b/grid_0.png

[Cathar]

Not gonna lie this is impressive. The items more than the characters, but I can see an use for both.


AI dungeon is precisely the reason why I am cautiuously appreciative of AI technology. It can generate instances where things make sense, but as soon as you ask the AI to follow up (ex "draw a dwarf mining some rocks" then "draw the same dwarf finding some gems in those rocks") the AI breaks down very fast. As for what AI dungeon has become...disapointment is an euphemism.

Generating pictures of artefacts can be actually useful to me, not to use as it, but to spare me the process of putting the elements together. I'll probably make some use of it in the future.


Edit : After a second look at it, it may be less impressive actually. None of the female dwarves is supposed to have hair, but the AI cannot understand the concept of clean shaveness. The items are very good looking but most of them do not depict what they are supposed to. This produce impressive looking pictures, but as of now, the technology is pretty much a gadget that cannot be seriously used on its own.

[neobit]

AI dungeon is precisely the reason why I am cautiuously appreciative of AI technology. It can generate instances where things make sense, but as soon as you ask the AI to follow up (ex "draw a dwarf mining some rocks" then "draw the same dwarf finding some gems in those rocks") the AI breaks down very fast. As for what AI dungeon has become...disapointment is an euphemism.

I feel you. Had the same experiences.

Edit : After a second look at it, it may be less impressive actually. None of the female dwarves is supposed to have hair, but the AI cannot understand the concept of clean shaveness. The items are very good looking but most of them do not depict what they are supposed to. This produce impressive looking pictures, but as of now, the technology is pretty much a gadget that cannot be seriously used on its own.

Not all of them were set to be shaven, but maybe it is even worse, as it looks like it completely doesn't care. It should be possible to add more context to make the \AI\ eventually figure it out, but I have failed to make it precise so far, even with additional information. Maybe it requires some more knowledge about how to provide good information for it.
Note that above only used words from DF. I added nothing to the description, just cropped it a little and added priorities. There should be a way to automatically extract from DF what is really important to AI to make it more precise. Or use the whole AI engine specialized more in characters than a generic one.

Eventually we will be there.

[BlueManedHawk]

I'm pretty pessimistic about machine learning in general.  I find that it tends to not really be quite as effective as carefully–thought-out algorithmic solutions to problems when such solutions are possible.  (Personally, i also find them to just be so much less interesting to study.)  While i won't fully rule anything out, i don't think that machine learning will really ever be anything more than a novelty and something that tech companies use to make themselves look cooler.