So a while back I made the castle: Fantasy castle - done! Today I spent a bit of time and used AI (Stable Diffusion) to improve it…
7 Likes
Oh boy, this is nice.
A couple of weeks ago, I did something with AI, by entering some nouns, and the AI generated an image. The given examples were nice to see, but I did things like “A car on a bike”, which gave absurd images. Fun to see, but worthless.
I forgot this site or the name of the AI software. I had the impression, that more of these tools were based on the same AI (software).
Everybody is becoming an artist … 
2 Likes
AI art is cool. I haven’t tried it myself yet. I’ve heard that it’s difficult to get exactly what you have in mind. But you did a really good job of recreating your concept. Did you use long and very specific prompts?
When I get the time, I’d like to try AI to generate some concept art for me to practice my 3D modeling with. I like modeling but coming up with the composition and everything can take a lot of time, so it would be pretty sweet to have it done for you.
2 Likes
Ask 100 artists for “A car on a bike” and see what you will get 
All of those are based on the same concept from AI research (diffusion models). There is one version that’s open source - Stable Diffusion. Many sites are based on that.
But 2 main competitors, which are better in some ways, worse in others - are not based on the open source version, they were earlier. One is Midjourney, one is Dalle2.
For the above - I did use the open source one. But now I have used the 2 mentioned for many hours.
IMO, the hype will die down. For some it will mean that the barrier to entry is lower so for sure there will be more artists (like it is thanks to Blender already). But maybe in 5 years it will be just a standard part of all the drawing softwares. But to be fair, I am not a prophet, so that just me guessing
No, not really. I did use those systems for a bit right now and I started to have a ‘feel’ what works and what doesn’t. And I did feed the AI my image to start with 
If you don’t have much time on your hands to learn AI systems IMO it’s better to use google image search or pintrest to get good concepts. AIs are (at least for now) not a low effort thing giving stunning results in 2 minutes that people claim them to be.
2 Likes
Yeah, I get what you mean. It’s not like you press a button and you get a stunning image in front of you straight away. You actually need to figure out what prompts to use and how to manipulate the AI to get something that looks good right? But the thing that fascinates me is the randomness. All I’m saying is that, after some experimentation, it can probably generate some fascinating scenes that would be fun to try to 3d model.
1 Like
Actually, sometimes you do 
And when you are just having fun, this is great. You put some words, press the button, get some results, some of them are visually stunning. Be carful though with that - it’s very addictive loop…
But when you are working towards specific goal, things get more complicated.
Also… it’s good to know what you are doing first, i.e., having at least good grasp on fundamentals of lighting, color and composition. I got nice looking images, that after closer inspection had impossible physically lighting, colors that don’t work together well or something was stunning at first, but after looking closer the composition was “wrong”. But to be fair… you get those things with human-made art too
Yes, and the most important rule of all is to have fun while doing it
1 Like
Oh ok 
But doesn’t that take away from the sense of achievment that you get from making art the “traditional / human way”?. I can’t see myself looking at an image and thinking, “wow I’m so proud that I came up with that word to describe what I wanted the machine to do for me”. Haha.
But I do think it’s very interesting and that it should be used as a tool rather than to be frowned upon and discarded.
How do you write prompts for:
Like what sort of syntax works? It would be nice to hear what you’ve learned so far if you’re willing to share.
1 Like
I think it affect anyone differently. One of the things for me, when doing 3D is that the process is fun. Also, when you think about it… when using Blender you are quite far removed from traditional/human way already - you are just moving your mouse and computer does most of the job .
And I’ve already seen people very protective/defensive about their prompts…
That was the prompt:
fantasy castle near the lake on the cliffs, single tower on the island connected by ((((stone bridge)))), (sailboat), a few seagulls, god rays, light shafts, volumetric clouds, by artgerm, greg rutkowski, golden hour, sunrise, cinematic lighting, high details, fantasy environment, medieval, volumetric lighting
I’m 95% that syntax doesn’t matter really. Just keywords. And in case of this workflow: source image. What I usually do:
- describe the subject (here you usually have to experiment, e.g. sometimes man with a sword works, sometimes swordsman)
- throw in some artist names - don’t think it will copy their style, it might for some - but in my experience it just shifts the image towards ‘style cluster’ (like drawing vs photo vs render); though maybe some artists have stronger influence than others
- adding style word on top of that (like artdeco style or cyberpunk style)
- add some words that might or might not improve things like ‘god rays’ in the above example
The words in ( ), are for more emphasis (that’s Stable Diffusion syntax for it). Stone bridge there is in multiple of them as it tended to remove it without it
Mind you, I mostly focused there on other parts (like how to tune SD parameters), and didn’t really put much effort into the prompt.
1 Like
True. The most important thing is to have fun. But I would be lying if I said there weren’t some tedious moments too. But that adds to the sense of achievment I feel after I finishing a big project. And moving a mouse is not so different from just moving a brush or pencil. Just different tools. And that’s how I see AI as well, another tool that I want to try to take advantage of at some point. I’ll have to try it first, and then I’ll have a better idea of what I think about it.
Haha yeah I thought so, that’s why I asked if you were willing.
Thank you so much for sharing! Very interesting. Now I kind of know how to begin. Will bookmark this post for my own AI experimentation later.
1 Like
What’s wrong with this?
AI impression of “A car on a bike”.
The funny thing is that every AI tool gives different outcomes on the same input.
1 Like
Check this one too: Understand PROMPT Formats: (IN 3 Minutes!!) - YouTube
And if ‘later’ will be in a few weeks… that might be outdated already 
2 Likes