This post will be a little different from the usual ones. I won't be talking about sinathir updates directly. Instead, I will talk about my experience with experimenting with AI and what it means for the game. AI is quite a heated topic right now. I will share some of my opinions on it, but I admit I didn't do that much research into into it and some things I'm going to say will most likely be inacurate or wrong. 

Also, if you have any opinions on this topic, don't be afraid to share it with me. Just please, be sure to read atleast the TL;DR sections before you do so and don't shape your reaction based on the title alone.

Introduction

Short introduction, I recently played around with different forms of AI. I don't hate it and I don't love it. As an outsider, the entire thing seems to be slightly similiar to the early days of steam green light or when unity engine was becoming popular. At that time, steam was flooded with a large number of cheap games made with stock unity assets, which required very little game-dev knowledge and efford to make. Many smaller developers were angry, because these "asset-flips" made it harder for their own games to be noticed. This also gave bad reputation to the unity engine and a lot of people hated on the engine simply because of the great number of bad games that were made with it. 

While that situation is not directly comparable to what is going on with artists and AI, there are some similiarities. I don't believe AI will kill all artists and take their jobs and I don't believe it will allow everyone to make perfect art with no skill or efford. Those with no artistic skills will not be able to use it to its full potential, while those who already know how to draw will only benefit from it. Ofcourse, I might be wrong.

I tried two types of AI. One focused on images (graphics) and one focused on text to speech (sound). In the upcoming categories, I will talk about my experiences and how, if at all, I decided to integrate this technology into my development workflow.

Character graphics

TL;DR: I won't be using AI for character graphics. Only exception will be placeholders or taking inspiration for shading or coloring.

Perhaps the most obvious use of AI is to generate images for characters. I tried stable diffusion through NMKD. It's free and doesn't require any complicated setup (https://nmkd.itch.io/t2i-gui). At that time, I had no plans on using it in developing Sinathir. I was just testing it out.

I didn't initaly bother with ordinary image generation. Instead, I tried img2img. I basicaly loaded an image and the AI was suposed to "make it better". After about an hour, I got this.

This is probably the best result I got. It passes what I like to call "first glance test". To me that means the image looks nice at first glance, but if you look at it for more than few seconds, you start seeing mistakes, errors and general AI weirdness. 

While this looks much better than anything I can create, there are several problems. First, the results are inconsistent. And from what I read, that isn't really something that is going to change any time soon. For every good looking picture, I got several pictures that looked like this.

Now she looks angry for some reason. Also, her eyes are wrong

Nika after few hours of sniffing glue

Those are results with same settings and same base picture and all of them are different from one another. This alone makes AI unusable for making main character sprites, since I need them all to be in the same style, with the same, consistent details. My art is not good. It's bad. But it is consistently bad and I think consistency in game art is very important.

It also won't work for making sprites for other characters, since style differences between the main character and side characters would be too great. Again. I want things to be consistent.

The second reason is much more simple. It's not fun. I don't enjoy writing prompts, waiting several minutes for the AI to spit out dozens of images and then sift through all of them to get something usable. I enjoy modeling much more, even when I'm not good at it.

I might have a use for this AI however. One potential use I can think of would be for making placeholders. Even drawing shitty figures, like Billy for example takes more time than using AI. And the result will be, without any doubt, better. I'm also not concerned about consistency of something that i plan 100% to replace with somethin else.

Another use I can think of is simply inspirational. I quite like the shading and colors on some of those pictures. I can simply try and replicate it in blender to some degree, which would make my models look better.

To close this section off, no. I won't be using AI to make character art. It simply doesn't suit my needs.

Enviroment graphics

TL;DR: I will definitely be using AI for creating textures. Actual objects and models will be still be made purely by me.

Now this is a diffrent story entirely. I don't like making enviroments or making textures. I still like modeling them, but I don't like finding textures and I can't really make my own. 

I want my game to have that early 2000 look. Something between alien shooter and stalker. 

I think using altered real life photos is a good way to achieve what I want to do. There are some problems however. Most of the good textures are copyrighted and not free. And those that are free are usualy not good. AI is actualy quite good in this regard. Being able to generate seamless textures on all four sides is a nice bonus.

I tried to generate a simple brick wall. This was the result.

Not usable for a first person game. However, my game is a top down RPG. And it uses 32x32 sprites instead of full sized textures. All the imperfections will be pretty much gone once I resize the texture and play around with filters and other effects. Here you can see different resized textures on a 3x3 grid without any alterations done to them.

I can safely say that I will definitely be using AI for making textures like this. It will not do all the work for me. I will still have to adjust them and edit them, but it will make my work a lot easier. Other than that, I will still be making the actual objects and models my self. 

Voice acting

TL;DR: I might use AI generated voices for the intro sequence and endings. Not decided yet.

Last type of AI I tried was text to speach. I came into it with no expectations and to be honest, I was blown away. (https://beta.elevenlabs.io/)

However, like with stable diffusion, there are problems. Again, the lack of control is a big problem right now. You can't change voice emotions or things like that. And the fact that it is not really free is another, even bigger problem.

Truth be told, I don't want my game to have full voice acting. I don't think it is necesary for this type of game. I would like to have it in some capacity though. Voiced intro would be nice, maybe even voiced endings. But nothing more. I tried to add voice to one of the early cutscenes  in a video editor. just for fun. To see if it ends up being as bad as I thought it would be. I was not dissapointed.

Reminds me of all those japanese games from mid 2000.

Don't get me wrong. This absolutely amazing. I can hardly tell that this was done by a computer and not real people. But as I said, it has the same problems stable diffusion has. One text line sounds different every time you generate it, so you have to generate it multiple times and choose the best result. Pair that with very limited free capabilities of this particular AI and you will realize that using this anything more than some short, important dialogue section is not really possible.

That being said. there is a chance I will use similiar technology in far future to make voiced intro and maybe endings. I can promise you I won't be putting it into the entire game.

Music

TL;DR: Didn't find the AI I wanted. Wish it existed, but it seems it doesn't.

To be honest, I wasn't looking that hard for this one. I found some AI music generators, but they either sucked or were not free at all. 

I didn't investigate this type of AI that much, because I'm not at the stage when I want to focus on music. I actualy like making music. I just don't like the tools. I tried FL studio and other free alternatives and  I just don't find it fun to work with them. I enjoy making very simple music with two or three layers, but turning it into something more complex is not  a pleasant process for me. That's partly because I didn't put enough time into those programs. I will revisit them at some point.

Right now, the game uses "free" music and I put free in quotation marks for a reason.

If you go on youtube and search for free music, you will find tens of thousands of results. Most of them say they are free for monetised and non-monetised videos, some are free only for videos without adverts.

The wording is very important here. Some music pieces strictly say they are free only for youtube videos, some say they are free for other projects and some say they are just free. Very few of them have actualy any info about the license they use and what are the actual conditions for using them. Are they free for videos only? Can you use it in other projects? Do you have to buy it or can you just download it? Who knows. 

I tend to avoid music with unclear usage rules, but there are some pieces that I use. I would like to replace all of the stock music eventualy, just so I can sleep calmly at night.

Now for the AI. I don't know if it exists, but I would like there to be a software that takes a simple midi tune and adds other instruments to it. Pretty much what happened with the only original track Sinathir has right now. 

I made this tune in https://onlinesequencer.net/

Then I sent it to my buddy, who was at the time learning to make music (he lost interest in making music completely since then so no more tracks like this.) and this is what he created:

If there is, or if there will be and AI focused on music, this is what I want it to do. Take a simple tune and make it more complex. If you know about anything like that, please let me know.

Last words

What follows aplies to all sections. This game is free. I'm not making any money from it and I don't want to change that. I also don't want to put any money into it, unless I think it really necessary. I can't hire artists, I can't hire musicians and I can't hire voice actors. I want to do as many things as I can by my self, simply because I find them fun, but making a game is quite a big process requiring multiple skills. Some of which I simply don't have and I most likely never will, and some of which I don't really find fun participating in. I don't find making textures fun and I can't voice a female character because I'm not one in real life. That's why I even started looking into AI. Not for it to replace things I like doing, but to help me with things I actualy don't like doing but have to be done anyway unless I want to compromise my vision in some way. My stance is prety much unchanged. Right now, AI is a tool, not a replacement. It is not a second coming of christ and it is not the end of the world (unless it gets drasticaly better in the upcoming years. Then we might have a problem).