I think this topic is a bit of a challenging one to handle. If AI is useful enough at accomplishing a task, it will eventually remove the need for humans to do that task. This has been the case with all technological advancements. Because of this I feel that it is kind of useless to grandstand. But I have wanted to talk a bit about the pitfalls with it and why I think it's both kind of "overrated" but also kind of not. mind you I made these assets in probably less than an hour, so I was actually kind of surprised at how quick it was to use. If you feel that it is obvious that this was AI art, and that it's terrible, I think realistically I could probably create a scene where you couldn't tell. I'm considering making a video where I use AI to create the best game art that I could, just to see how it stacks up, but I also have a lot of other topics I want to cover, so we'll see. As always, feedback is appreciated. Thanks for watching
I think you handled the topic very well! I think that treating AI/generative models/machine learning like the boogeyman ultimately does a disservice to artists because true empowerment will come from knowing how to deal with this tech, rather than leaving it to ignorant profiteers and greedy corporations to do what they will with it. Of course the ethical concerns of how this tech is being implemented and used are important, but they aren't the only valid point of discussion. Treating it like a serious threat also runs the risk of "validating" the idea that AI "art" is on par with human creativity, when the reality is that these models have no real understanding of what they're spitting out, and this will always be a limitation, since art isn't just aesthetics - as you pointed out, the artist's intention communicated through their work is just as important, so is the creative process itself.
The biggest problem with IA tech is that it only works if the majority of people don't use it, let me explain, what IA does is essencialy turn images into digital noise and then rebuild it, it does that with a billion images and boom you get the product, but you might have notice a couple of problems if the image especialy when there's repeting patterns, the IA kind gets weird with them, and not only that, but a lot of meaningless stuff just pops out, like the clouds on you're example, now if you imagine IA using IA images to train itself, wich is gonna happen if enought people start using it, especially if it's used in games and so on, them those little digital anomalies will start to happen more and more, this happens with any program if you make it interact with itself enought times, I'm a physics student and I worked with research for a period, this is a problem that happens a lot in physics research we always have to take a lot of care so programs don't start generating crazy results, it's acctualy realy easy to make the computer go crazy by making his output into input, in that sense is like inbreed children the problems with the DNA start getting bigger and bigger until the whole thing falls apart, and take into consideration that happens in scientific reserch the top of computer programing and data control, so IA's like DALL-E or midjourny are gonna start to become completely uselles with they become the main source of art, imo what's gonna happen is that the outputs are gonna start to get weirder until they become bassically just noise and the program can't turn it into anything, but it can also explode or smth, we really can't tell with this many variables, but if it starts getting it's output and reusing it as input the program is gonna fail, that' just the reality with computers, even if we think about them as perfect calculators in reality they can only count in 1 and 0s everything else is built with ducktape and prayers, and if you need anything with a lot of precision you're gonna need a LOT of prayer
I feel this is a fair assessment. AI always felt to me like the false promises of a "silver bullet" to not having art skills and being able to avoid having to actually put in the work. The continued devaluation of artists and the value of art was always there, it's just gotten more depressing nowadays with how obvious how much care folks put into quality.
Everyone familiar with the techniques have since the start - almost two years ago - been clear that the tools are best used by artists who also know how to make the most of them. It was never meant as a silver bullet. I think that is people mistaking it. Skilled creators can only rise higher with better tools.
I've been feeling like google is terrible at references for quite some time. I mostly do nature assets and stuff so I tend to go to botanical gardens and similar and just take my own references.
Thank you for using the correct graph to describe the Dunning-Kruger effect! It's equally funny and infuriating how many people very confidently explain it completely incorrectly with made-up graphs.
Great video, I'm a professional animator in movies and I've had people ask me about ai potentially threatening my job as a cg animator, but even if they could get it up and running, they will lack the polish and the eye that comes from doing it for 10 years. With this being the case, do you think it's better for someone to ignore ai illustration completely and learn to develop skill and taste through hard work and intentional practice? Or would it be better to resume with ai and develop that "taste" you likened to interior design? And if you prefer the latter, how does one develop that eye and taste?
I mean as someone who likes to draw and will continue doing so I feel like it's tricky developing taste without actually understanding the craft at all. I suspect that the best course is still to learn how to draw and that that will stay true for a considerable amount of time. I'm just not entirely sure it is up to us, it is up to the consumers and the companies. Knowing how to draw will continue to be valuable, but I expect there will be pressure to use AI in order to maximise output.
@@Nonsensical2D Thanks a bunch for your answer, I want to learn to draw but there's a sense of futility with the impending ai programs getting better and better. Good to know there's artistic cultivation that mostly happens from doing, and even with a pressure to use it, drawing will still bring its own fulfillment.
@@MischiefMaker-mk9wn If your desire to learn to draw is driven by the potential money you plan to make with it. It probably won't motivate you long enough to get good. If you actually enjoy the process of drawing, feel the internal reward of creating something, of manifesting your imagination to the outside world - then it wouldn't matter how fast a program can generate pixels that it doesn't understand
Great video. The problem is that when artstyles of assets don't match, the game doesn't look good. These algorithms are not precise by nature, at least not the ones there are right now. So it's super difficult to let it create a cohesive set of assets. It's not a matter of "just train it longer on more data", it is just not in the nature of how these algorithms work.
Haha, I considered using the "other" one, the incorrect one, because it is slightly better at illustrating the effect, but I knew there would be people out there (like me) that would be annoyed by that graph.
Thank you for making this. It's a touchy subject and I've long wanted to hear from an actual artist who doesn't just immediately put the blinders on and scream about AI stealing your lunch money! I'm bad at art and frankly I'm ok with that. I'm a programmer with a ton of hobbies and I'm ok with not having this skill (or being musical). If I could snap my finger and be a great artist and musician I would obviously do it, but I can't. One of my hobbies is games (video and board) and having the ability to use AI to generate music and art is a great thing for me. All the stuff I do is for me and a few friends, so I'm not making any money on this. Hiring an artist or musician to generate this content (and it is a LOT of content that would takes dozens of hours of a human's time) is not something I could ever afford to do. AI is not perfect but it makes the things in my head possible when they wouldn't be otherwise. Two styles I often use is photo-realistic and plastic-miniature, which is not something most artists would do anyway (I don't think). Even using AI, I've still spent well over a hundred hours tweaking prompts and manually editing the results to get closer to what I'm looking for. For what it's worth, I don't think AI will replace actual artists. It's too hard to get consistency. AI can be used to quickly generate "themes" or what not, to allow a game designer the ability to clearly express their vision, and then pass that off to artists to bring it to life. Same with music. But you'd need a human to keep everything consistent and put in the finer details and love. My biggest fear is that AI will lead to even more overcrowding of media. We already have more shows/movies/music/books/games than anyone can possibly consume in a lifetime. Indie devs pushing games to Steam mostly just get lost in the flood. AI will make this 1000 times worse, with AI garbage art/music/programmed content being slapped together on what can just barely be called a finished product and then shoved out the door by opportunists looking to make a quick buck. For what it's worth, I always thought the future of AI was to do the boring repetitive tasks and free up humans to pursue creative works, so this AI art/music/writing thing completely took me by surprise and I'm still kind of trying to wrap my head around it all. If AI is doing both the boring stuff and the creative stuff, what do we humans do with our time? 🙃
it's also worth noting that in their current state, generative models like "AI" image generators and LLMs are hugely inefficient, with every single prompt generating a disproportionate amount of carbon emissions compared to any other programming tool. it's an engorged version of your phone's text autocomplete suggestions, and unless you're curating a specific model from the ground up yourself, it's fed off of the labor of other artists without their consent or compensation. generative models definitely have uses, they're a powerful piece of software and have many practical applications , but they require stringent human moderation, are overhyped for what they are actually able to provide, have an incredibly destructive environmental effect when used at this scale, and honestly shouldn't really be called "artificial intelligence" in the first place. they're an early form of an imperfect, but potentially useful, set of tools that is currently being used excessively and poorly.
I had a similar problem with AI...I could not get it to understand "Isometric" Every attempt had perspective. I like your channel great list of videos👍
Can you show us your drawing process start to finish like making the canvas on procreate so we can see the size and then how large you make ur asset and what brushes you use I struggle with sizing without implicit references
The video really hits the nail on the head - I've been trying to work with AI to see how it can help me with my game art and have found it pretty useless so far, which definitely makes me feel more secure as an artist. The only real use for AI I've found so far is to create targeted reference images for my hand-drawn art. But I also agree that, while as AI develops it will be able to produce better art, there will always be a place for people who understand art techniques and have an eye for these things, or as you call it, taste. I've always thought that certain people are natural artists because they're able to see things in a way others cannot, and (we can hope at least that) AI is not capable of this. Or perhaps so far AI has mostly been trained by non-artists who also don't have that kind of eye?
We feel validated and acknowledged. ( I'm a 2D game artist myself. ) You described the unexplainable pain of artists so precisely. I had a war inside me about whether I should keep making art as a career. Your last sentences give me new hope. Thanks!
I never plan on using AI since it can't give me the unique things I want. I tried getting AI once to make a picture of a brontosaurus with gorilla arms and a giant eyeball for a head and....yeah....it gave me a vague dinosaur with like 3 1/2 eyes, etc. It was not at all what I wanted. So, if you want unique assets then humans still need to be involved. I've heard people say it's good for programming so I tried to get it to make a spawn script on a timer and it screwed up a mere 4 lines of code. I still had to figure out the solution myself even though it did put me on the right path and I eventually got what I wanted. But that's leagues away from, "AI will make entire MMO's in minutes" claims.
Cool video, I am scared of AI as hell. I don't think that a non artist can produce cohesive scenes efficiently enough. But with enough research I do believe that artists can actually use AI as a tool efficiently. Combination of drawing for AI to well define the artstyle with more experienced prompting I do believe that it is possible, and that is exactly what scares me most. Artists would eventually become pressured by the time requirements and will start using AI, and that could start the downfall for all of us. I am really hoping that AI usage is somehow legally managed, but I am afraid that train has left the station long time ago.
I think the best uses for AI for artists is actually to make procedural repetitive stuff, then using that stuff to make your original art better. (Stuff like generating rocks texture for a 3d material, or to use with blend modes on photoshop, generating simple seamless patterns) Because some procedural stuff actually looks more natural when randomly generated than doing it manually. But when it's something that you know what you want, it's super frustrating to try to micromanage the AI to make what you envisioned, so it's far better and more fulfilling to make it yourself. (And sometimes it's even faster than the whole prompting process) That being said, I think you should try out Krea, it's pretty cool for testing out compositions and lighting. You can draw just the silhouette and see how it might turn out when it's fully rendered.
So I am attempting to do a 2.5D too down game. A cozy style game. I want to draw everything. How would you go about drawing buildings for the town. Thanks. I subscribed because you actually teach. You are wholesome and honest with everything. Thanks!
There's this relatively older technique which is style transfer. A few years ago it was getting to the point where you could shade a ball and it would adjust a scene to the style in real time. I feel like all the full-on generative stuff kind took the whole spotlight, but style transfer is something much more interesting in my opinion.
its a bit tricky. I talk a bit about it in a video of mine called how to learn game art, there are some courses like Drawabox, but I personally think just practicing and analysing gets you really far, especially if you want to do it for game art, and not general art. There are a lot of skills that are significantly more important for game art than they are for traditional art, so I think practicing normal fundamentals while helpful might not be optimal (and probably less fun).
A criticism of text prompts as an interface comes to mind, which is that it makes visual art(which is driven by our senses) into a social construction(language). We do this naturally in the process of drawing, to break things down and simplify them, but that process puts the artist "in the driver's seat" as this video aptly describes - making new language from the combination of the symbolic understanding and our senses. Prompting omits this basic connection. I believe interfaces that act to extend or stylize sketched drawings fare a little bit better, while still losing the information communicated when the artist is doing the rendering.
Do you feel like creating the assets in grayscale then doing your own thing with them would save much time ? I don't really like AI art in general and the moral implications but I have thought about using it as base art if it could speed up my game art creation some.
I'm honestly not sure, but probably. At least for assets that would take a lot of time to make. Cause you generally have some assets that take 3 hours to make and reducing time on those assets can be a huge time saver. For other assets like tiles, I personally feel like AI is quite useless, the prompting becomes annoying and it takes me 20 minutes to make it exactly how I want it anyway
hei im new to game desind can i ask what you use to desind your assets and what you use to put then together and making a game map and also i watch your video where you make runing animation using key frame what you use to put all the keyframe together. you also but your assets, make maps and then add the running animation into it can you tell me how or please make a video about it thankyou verymuch
would be really helpful a video where you show your process from creating an asset to importing it into Godot. And also how you use tilemaps in Godot to structure your level with hand drawn arts. Thanks a lot!!
This is very much my experience. I've had some luck with getting quick and easy placeholder art for some projects where I was only interested in the code part. But it had exactly the same problems that you ran into. If you have a slumbering art interest it's honestly way more rewarding to create your own art and music. Aseprite is top of the line for pixel art and really cheap, and a midi keyboard and a free DAW like Waveform from Tracktion + some free instruments offers you music creation capabilities recording studios could only dream of back in the day.
I used to think AI art is also art but i've done a 180 on that take. Art is deliberate, there's thought and meaning behind every minute detail. There's no art without an artist and AI is not an artist, it's just that, an algorithm. I hope that AI doesn't improve further that CEO's of AAA studios lay of their artists in favor of AI to chase more profit. It would suck ass
I think the question whether 'AI art is art' doesn't make much sense. As you said, art is deliberate. AI is a tool, and if you play with it enough, you'll know that lazily prompting doesn't give you the visualisation of the idea you want to portray. The art-iness comes from the person, and when people say the images look bad, it comes from the person being bad at it. Asking whether 'AI art is art' is like asking whether ink being splashed on a wall is art. It might or might not be. If you just bump into something and spill the ink, that's probably not art. Trying to define that all AI generated images either is or isn't art is misleading. And regarding CEOs laying off artists, I don't think it's about the improvement of AI. That's the vision (or near-sightedness) of the CEOs. Generative AI is a very powerful tool for human creators, and it should be improved to the fullest of its potential, while still be ethical.
@@kieselzusammen This is a quote from a sculptor I saw in a video once "It can feel a lot like magic, because when I move my hand, I place my chisel and when I swing the mallet I know exactly how the stone is going to break. It doesn't make any sense, but I do. I visualize it, and then it comes true." AI algorithms, among other things, are dependent on data sets. They're not made out of anything as part of some natural process, that follows natural rules. They're not stone, they're not even ink. Tools are instruments wielded with precision, used for craftsmanship. Imagine you had an algorithm that could remix cooking recipes. To me calling the conveyor belt of data that remixes art "tools" is like calling the recipe remixer "ingredients." As for CEOs "ethics" is not in their job description.
@@futurestoryteller Tools are anything that make your work easier, including automation. AI can also make precise images, even without you drawing over parts of them yourself, but you have to be very precise on how to prompt and use other extensions to do it. So it's just not that good a tool yet, but if we don't count this a tool, then editing photos with some Photoshop tools like Content-Aware Fill doesn't count either. The remixing of data is also kind of inaccurate. Many AI models work by denoising randomised pixels until it fits the patterns it has learned, but it's not exactly just mishmashing the original works. If we think about how we know what the correct highlighting and shadowing should look, we don't work that much differently. We saw a lot of patterns of how light and shadow should look, and we put it into image in the way that should make sense in the context of the thing we want to portray. The difference is AI doesn't understand context. It doesn't know what makes sense, but it isn't necessary for it to do. Being a tool means it's already okay if we let all the context-related considerations be the user's responsibility. Actually, that's how things should be. You shouldn't go 'look ma, no hands' when you try to create something. The main reason I disagree with the notion that it just remixes art is how it's impossible to store the amount of training data those models use within the size that they really are of. You can't combine two things together if you can't access both things. Those AI models can't reproduce the original works they are trained from. (Technically, they can, but only when they are 'overtrained' which means something is just too prioritised that they ignore other training data, so the vast amount of original works still couldn't be reproduced.) Everything is just reduced to parameters and equations representing what it has been trained on.
@@kieselzusammen A person who does all my work for me would make my job incredibly easy. If I've done that, am I the real artist, and the person the tool? Saying we simply recognize patterns and copy them is a reductive take on human consciousness, if not the skills necessary to make art. Computer graphics software can simulate shadows and reflections better than humans, the only reason we don't pretend they "understand" the data we've plugged into them is because they're not marketed under the pretense of simulating actual neural networks. They are not at the user's discretion, as demonstrated by this video, the fact that it will never actually understand what you want, while it relies exclusively on what you tell it (as well as what data it's collected) means you are always at its mercy, not the other way around. Those parameters and equations are the data it's remixing. That reduction of size is a red herring. I'm still pretty sure you're taking something that's a more complex version of something completely normal for computers, and because of the complexity with which it operates, kneejerk ascribing virtually anthropomorphic properties to it. Image and video files are compressed all the time. But because this time it's complicated enough to _appear_ abstract we're just accepting the facsimile as "authentic" to something other than normal compression. A good example of AI proving this concept in practice is "deblur" and "depixelate" AIs. If you have blank film frame a normal compression algorithm notes a pattern and reduces the number of pixels it deems significant to reconstruct the overall image, right? AI is doing that, but for more complex patterns of pixels. The fact is it's just doing what computers have always done in a fancier way. It doesn't see or understand the outputs. It's still just made up of numbers. It's not an art generator, it's an art calculator.
@@futurestoryteller My point wasn't 'people also copy patterns,' but 'AI doesn't just copy patterns.' It is clear you think AI does that, so when I compared it to people, you thought I meant people just do that, which is the opposite of my point. > They are not at the user's discretion, as demonstrated by this video, the fact that it will never actually understand what you want, while it relies exclusively on what you tell it (as well as what data it's collected) means you are always at its mercy, not the other way around. I'm not sure how you think AI as a tool is supposed to work. Should it be self-sufficient? Then it wouldn't be a tool anymore. Other tools 'rely exclusively' on our control as well. Currently the most common way to control the AI is just through language, but it isn't the only way you could do it either. The example of the tree in the video might be able to be done if you can use some extensions like inpainting. > A good example of AI proving this concept in practice is "deblur" and "depixelate" AIs. If you have blank film frame a normal compression algorithm notes a pattern and reduces the number of pixels it deems significant to reconstruct the overall image, right? AI is doing that, but for more complex patterns of pixels. Except that isn't how it works, and I doubt if they do that, we'll get the same results. First, suppose we want to reduce the size of the original works, we can do it lossy or lossless way (meaning if we lose any quality in the data we store.) Lossless is out of the picture. We don't have an algorithm that could reduce the size that much yet. So if that's how we want to do it, we'd have to store the images as the lossy ones. And the problem with reconstructing the lossy images into crisp-looking ones is how, as we know, machines are dumb. If we just let it guess how a blurry image should originally look, there's almost no way they can get it right, especially with details like human face. It needs to know how high resolution images look so it can remake the low-res ones somewhat convincingly. Now the problem is recreating itself, because if it only stores reduced quality images, there's nothing for it to compare with so it could reconstruct how the image should originally look. I'm not under the impression that it's sophisticate enough I anthropomorphise it and say it does things like humans do, but your speculation on how it works couldn't be correct due to inherent technical limitations. And if we get abstract enough to say that even if it finds out patterns and store mathematical parameters, it technically is just compression, then this is so broad that any kind of learning, even human learning, could also fall into this category. Because we still have to store what we learn physically. Researchers can even reproduce images from our memories, even though in very specific conditions and with poor clarity, so we also 'compress and store' what we learn. Again, the point isn't 'it works like humans' or 'humans just remix from memories too.' The point is 'what it does isn't just remixing what it stores,' because that doesn't work. > It's not an art generator, it's an art calculator. I completely agree with this.
I tried to use AI for creating assets like rocks, trees. However the style and perspective does not feet my expectation. As you said, it is promising, but from my point of view it is still not ready for helping. The one way I see it can be used now is just for creating some reference, or creating drafts. Stacked with creating background and levels. Before cooperating with a human artist, I tried to use AI to understand how it should looks. No results =(
The only time I find AI art anything close to useful, is just for a moodboard or first draft concept art. Like for instance, maybe I meed a monster. I prompt midjourney to give me an image of a monster. Then draw my own version of that monster with better proportions and details that are easier to read, then animate it. On one hand, it can potentially get rid of the creative block at times. On the other hand, I tend to get my ideas easily without having to spend hours on any of those tools.😅
You struck a chord with me. I have no idea how to identify identical artwork, I just know some things are off when I see it. Now I wonder if buying an asset makes anysense at all.
That trick of having the AI generate greyscale mostly except maybe if you are in the idea phase, then coloring in engine is pretty great. With the example you gave for AI art the worst part is probably the ground for the platform. But if you are going somewhat high res but without too much detail anyways. You could always just lay down some paint brushy gradients and use that. And then use the AI for anything that takes a lot of detail. the example you gave of simple hand-made art looks really charming though. And for the right game would look way better while being fairly easy to do. I completely agree it is pretty much unfeasible for the character stuff. Although cartoony character ontop of a more painterly or detailed background is a classic combo that works, so if I was going for AI on a game. (which I am not.) Doing something like that would be sensible. Overall, I feel like painting skills matter a lot more now that AI is a thing. Drawing also matters but basically only fundamentals like construction and not much more. Being able to block out rough colors/areas is valuable. Since in the context of a game, the assets themselves have to be used in context anyways. And having someone think about the role each asset plays is pretty important for making the assets work together
Most likely, to create actual game art, you'd probably want to use more advanced Stable Diffusion workflows where you can actually create finetunes or embeddings to force everything you generate to have an unified, common artstyle based on a selection of examples, either generative or real.
AI usually comes with a nice expressive palette. The art presented as human made on the other had has this low contrast moldy fungal feel. But it somehow got popularity with the games like Cataclysmo. Apparently indie artists now go for shock value, in both form and substrance, instead of pandering the spoiled consumers with nice aesthetic pieces. Then again, AI was explicitly trained not to produce revolting palettes
This is actually the exact reason I've given from the beginning for why AI can't make art, in any literal sense, because an artist has discretion and a neural network doesn't understand the meaning of the word. It can spit out seemingly infinite variation on the definitions that it's culled from the internet, but it doesn't actually _understand_ it. It's like if NASA trained a computer model on information from every space mission. It would be able to dispense fixes for common problems, or fix bigger problems that NASA has probably already implemented solutions for, but it wouldn't be able to understand a truly unique problem, something outside its usual data set. It's not thinking like a human
The web based interfaces such as Dalle are essentially just toys for one off images and useless for this sort of productive work, sort of like RPG Maker isn't what you'd use for building a first person 3D game. You'd want to get specific models and likely finetune them to reliably produce the style that you want, or use the various guidance techniques such as ControlNets. You'd also generally spend a lot of time either taking it over to Photoshop, Affinity, etc, to do touchups or change the direction of things before bringing it back over, it's very much a back and forth process. Some interfaces like Invoke seem like they'd be better for working in place without switching to other tools. You'd also want to be careful about which model you pick based on the image compressor/decompressor it uses, e.g. the 4 channel VAE of Stable Diffusion 1, 2, and XL can't do flat shaded artwork very well, but the newer 16 channel VAEs are great at it.
Very interesting video, it kinda put into clear words a feeling I had for a while with AI generated content. As an artist myself, I'm fairly interested in the matter, mostly from a distance, but still, interested. If I were to make a video game, which may or may not happen someday, I feel like using AI to generate bases for some assets or some variations of existing design then repainting on top of it, to clean it, add that "intent" and such, might be a decent option, especially if tackling a project like a video game on my own. Not sure I will, but, good to know it's possible and doesn't look too bad.
Well, yes, AI is no more than a tool. A REALLY multimodal tool, but a tool nonetheless, you need to have skill in the area that you are using it for to guide it correctly. I use AI to aid me in programing, and it solves a lot of things, I've learned a lot thanks to it too, but if I did not know what I was doing I could not obtain anything coherent from it.
You don't even need to be good at art, just to have a basic understanding of composition. I also recommend against it because it's a swamp of copyright issues.
Im not a good artist. But i can take a photo or ai image and make my own art based on it. But i agree that just directly using ai assets could be difficult
My biggest problem with ai art is how artificial it looks. Sure it's very detailed and looks professional, but it also looks really dead. I honestly would much rather see someone be bad at making art than to see a "well drawn" ai piece of art, because the bad art of something like an anime character or a cartoon character at least has heart behind it. A machine is not alive and only can make art through the prompts it receives; therefore, ai art just looks souless in comparison. It doesn't mean I'm threatened by ai art or I hate people that do ai art for fun: far from it. I just prefer seeing someone better their skills, rather than using the easy button.
it just copies the stuff the companies stole from artists. It has no intelligence, it doesn't think, its just a probability algorithm, basically it auto completes your sentence with a picture or in the language model with text. AI imo stands for artificially inflated. It will be used as a replacement for stock images, or to create mood boards when the bubble bursts. I already see companies in the game industry that struggle to finish their projects because they thought they could replace their artists with ai prompters, now they rehired artists to fix the bullshit that the ai creates but somehow still insist on having ai prompters because saying you use ai gets investor money...
Yeah there were a few weeks where everyone was enamored with AI art online. But after that it was like eating cake for every meal. People got tired of it and the groups I am part of started banning it.
In my opinion AI is a good tool to support artists, but a terrible tool to replace them. AI can generate little environment things and other side stuff, but AI messes up, A LOT. Having an artist to detect and fix up those flaws is necessary if you don't want every aspect of your visuals to have some thing that just looks... weird. Otherwise you end up like Once Human where, if you take a closer look, assets are SOOOO missmatched that it's honestly baffling to think about.
It's refreshing to see an artist have a realistic take on AI art without instantly dismissing and moralizing it. Overall, I agree with your main takeaway, although you might have been a bit limited by only using Dall-E. The process of simply generating an image using only the prompt and copy-pasting it into the game is deeply flawed, due to stylistic inconsistencies (that are going to be instantly noticed by the player, especially if it's a foreground sprite). Perhaps a more controlled approach could work, by using a local Stable Diffusion model and processing a rough handmade sketch through a pipeline i.e. sketch -> ControlNet/img-to-img -> inpainting-> background removal -> manual editing. I also think that using a specific model or model finetuning (e.g. Lora) to achieve a specific style might help in providing consistency. Maybe using AI art not to create base art, but variations of existing sprites could also be a better use case. For example, different customization options for the character or types of vegetation for different biomes. Something I would really like is AI to be able to create different poses and compositions of input character sprites (for animations too), but from what I have seen and tested that does not seem to be possible currently. As it stands, AI image generation is a tool that is greatly overrated by some people and underappreciated by others. But I don't think it is capable of surpassing us in the near future and completely replace us. And I sure as hell will not let AI take away my passion for gamedev and drawing, regardless of whether I make money with it or not. If it comes to that, I'd rather become an advocate for economic reform (such as introducing UBI), than to demonize and shun everyone who makes use of technological advancements.
Would you be willing to retry this experiment with Stable Diffusion's advanced tools? A lot of the issues you faced might be fixed using features that Dall-e doesn't have (eg. using inpainting to fix pieces that would be great if not for a single detail, use scribble controlnets to construct the basic structure of the asset before generating, use seeds to create slight variations of assets, or even train a lora based off your own art to train the AI to follow your style, which is something you said you were interested in). I agree with your statement about how much composition is important to art, which is why I'm especially interested how an actual artist with a trained sense of style can use the tool in a production pipeline.
Interesting. Ye, I could look into it, I'm a bit unsure how well it could run on my computer though (I only have 4gb). But I also think you could get significantly better results with dall-e than what I produced here, if you spend some time with it.
@@Nonsensical2D One of the big issues I found with Dall-e is that your access to it is filtered through Chat-GPT, meaning you type your request to Chat-GPT, Chat-GPT creates a prompt based off your request ("Make a game asset of a bush with a simple artstyle" becomes "game_asset, bush, simple_artstyle" or something), and then that prompt is shipped to Dall-e. It makes it really hard to experiment with since you're not 100% sure what it is actually being requested at any given time, and your response may or may not be influenced by your previous requests.
I've been telling people who say 'AI art is crap' that it's not about AI, it's about the knowledge and attention to detail of the person behind the AI. If you think the images look bad, it's often because the person doesn't know why it's bad or how to fix it, rather than because AI can't be used to make decent looking images. But my voice lacks weight because I myself don't know art very well. Glad to see your take is rather close to mine. Anyway, even if you lack the skill to draw over the bad stuff from the AI yourself, you can still draw over it badly and then make the AI use it as a reference to fix what you drew. I'm not sure if Dall-E can do this, but Stable Diffusion can.
its similar to when someone just takes a photo and traces it or comor picks. without art skills it looks really bad. same thing with rotoscoped animation without knowledge of 12 principles of animation - it always look bad. same thing with just copying code from stackoverflow/ chat gpt. you need to know what you are doing even if ure just supervising it
AI images look really impressive until you've made your first asset by yourself. No matter how crummy it might look, it'll be and feel 100 times better.
Wow. This is exactly what I felt when using these prompt made art AI programs. I’m a game dev and also draw art, but even if the AI art is “good” sometimes it feels off. So in the end you need your own skills to actually use it in your game, if you want to do that. Never late to learn new things I guess. Thanks for the eye opening video. It was great to know somebody actually having the same thoughts. I guess the road of art and game art is a tough one. But we should never give up keep trying to get better at art!😊
I agree that AI is inevitably here to stay. The worst part of AI for arts (graphical, musical, written or any creative art) right now is how it is trained by stealing the work of other artists without consent and giving no compensation for their work. All that said, AI is a tool and if they correct the legal and moral part of training it with other artists art, it can became a very usefull tool. I totally agree on that I prefer art created by humans with intention and meaning and I am also totally bias because I enjoy drawing. It is the process of creating art what is the most important, more even than the result itself. I also sadly agree on creative skillful artists will have to adapt to this new AI world. But at the same time, I think, and hope it will be a growing number of people that not only prefer but only want human created art. And I tottally agree AI is not a golden ticket on any field. If it is not guided by an skillful person that understands the specific task at hand it become mostly useless.Thanks for the video.
Ye, I actually somewhat speculate that artists that create paintings (like oil, acrylics, watercolour) might be somewhat unaffected for quite some time. At least when I've sold personally, 500+€ are sums where the person buying often want your signature and your craft to show through, otherwise they would have bought a poster. (though that side of art has always felt like a tricky to make a living).
The more we look into it, the more we realize AI isn't really intelligent at all. Calling it "AI" is a marketing ploy and a misrepresentation of what it really is.
Ye, for sure, I think I make a point to use the term generative AI (or LLM). I think at this point when discussing intelligent AI we just have to default to using the term AGI. It's kind of similar to the euphemism treadmill, but a kind of 'marketing treadmill'. Which is why you also see some people like Linus (LTT) use the term ANI (artificial narrow intelligence) to refer to OpenAI-stuff. It is what it is, but I agree.
I mean I don't really like AI art, I don't necessarily understand what AI game dev actually means, but I'd probably not like that either. I'm significantly more of an artist than I am a game designer, so AI art is actually a bad advancement for me, but It's not like I complained when digital cameras or phone cameras became a thing, so it'd be kind of hypocritical to complain if technological advancement does successfully replace me, it's not a world I'd prefer, but it is what it is.
@@Nonsensical2D I get what you're saying but when digital cameras came into existance they didn't create photographs by stealing other peoples work... regardless how we all feel about how AI looks at this point I just can't get past the fact that it's stealing from people like you and me (I'm an artist as well and I've done a few games). It's not really about being an uptight artist, more that our work and labour is being stolen and exploited while we continue to be underpaid... it's a bit different than technological advancements like cameras because they weren't built by scraping, u know? I learnt a lot about assets from you and I hope we get to learn more about creation in the future from you and not AI. Obviously do what you'd like but idk its sad to see so many artists take this direction lately on youtube.
@@lesflaya I mean I agree with you, and I'm personally not planning on using AI, I like to draw, it's what I do. I frankly don't like this development, but I also think the battle is futile, the only people that care are the ones immediately affected, the artists. The rest of the world just want a good end product.
@@lesflaya While many current image generating AI models were built on scraping, I'm not sure if it's all of them, and the way they work doesn't inherently need scraping or unauthorised use of images as training data. Some people also already made checkpoints for Stable Diffusion using only authorised and public domain images. Those checkpoints simply aren't that popular yet compared to the questionable ones. Would you still feel the same way about the models that don't steal from anyone? Because I personally believe that sooner or later, they will become more mainstream and might be able to replace the models built on scraping.
I have to disagree with your point, especially your point about the Dunning-Kruger effect. I don't have to be a musician to tell good/harmonious music from bad or music produced by an amateur or beginner. Same with art. I am no art critic, yet I can distinguish a 5 year olds drawing from that of a professional painter. Of course there may be an expert level where only experts and artists themselves can appreciate certain techniques or flourishes to the art. But that's not the main demographic of ... well, anything. I think A.I. art won't take over, it's not smart enough. However it will equalize and liberate people. Sure, bespoke and custom art will always be better and allow you to realise your vision with pinpoint accuracy - for a price. On the other hand even hobbyists will be able to create something with A.I., though the trade-off might be that it won't be as pretty and cohesive as with bespoke art.
I think the problem arises not when you have to distinguish a 5 year olds drawing, but being able to distinguish something good from something less good, or something good but which contains something that is "off". and also knowing how to handle it. Suppose you have something that looks nice, how do you know how much nicer u need to make it until it is 'just right', how do you know when to make something less nice in order to increase cohesion. This is a problem that might exist less if you use AI art to create an individual art piece, but when you are making a game you have a whole host of issues that you have to deal with, because you don't have one art piece, you have 500 assets that all need to go together
Your argument ignores a lot of nuance in what we see as the end user versus what went into creating that complete picture. Humans are very good at interpreting whether or not something is "off," but without training, we don't have the language to describe what's "off" about it. I think that was the point of the video. Like Nonsensical2D's comment mentioned, you're not feeding high level information into an AI. At this point in the development of generative tools, you're having to describe, in detail, how you need something tweaked to get a good end result. If you're using generative AI for your game, you're essentially occupying the role of Art Director on your project, a role that requires decades of professional experience to communicate with the nuance required to achieve a good final result.
@@calicow And that's where I disagree. I don't need decades of experience to see if something looks bad/off and yes, not everybody will be able to create coherent and good looking art, even with support from a.i. or other tools. However there is a wide margin, even between professionals. I would guess A.I. will, in the coming years, replace low skilled up to medium skilled professionals while simultaneously liberating the market and opening it up for a lot of semi-professional developers using the technology to create something. This is similar to what self-publishing and ebooks did for the book market. Is there a lot of sub-par low quality stuff? Sure. But on the other hand there are also AAA games and traditionally published books I wouldn't play/read if there was nothing else to do. This is just another tool in the belt. I, for one, am no musician. So if I can just pump out a passable tune with the help of a.i. then that is more than I could accomplish myself and if it's something for a small scene, that might be all I need.
I disagree with the central point of your thesis, which is also speculative in nature. You dont really know that people who cant draw dont have an eye for a creative process. To make an example; I cant sew anything, but i know how to dress nicely and properly. Same apply for the overwhelming majority of people who know how to dress. i'm extensively using AI and i can see when something is not working and doesnt fit the style of previously generated assets, despite the fact that i cannot draw. It seems to me that you missed the inconsistency of the point you were trying to make: you said that when people look at AI generated collage they can see it's not deliberate therefore less aesthetically pleasing. But who are this people? are they all artists? obviously not, everybody can discern good art from bad one, although it definately takes experience to discern from very good one to just good. The example you pointed out are just bad use of AI, mostly because it's an incredibly frustrating endevour, sometimes the AI dont listen to even basic instruction after 500 different iteration of the same thing, and so they might resort to give up and try the least bad attempt and mesh it together with the rest, but that's just the same as any other aspect that revolves around the creative process. Many art critics couldnt do anything even remotely close to what they criticize, yet they have the eye to discern if it's good or not, if your point was valid that wouldnt happen. I'v seen insanely good AI projects; there was a test done times ago where asked people to discern from AI images and real, and nobody could. there was a painting competition few months ago where an AI generated work won over "real" artist. Personally AI is the only tool that i can use to generate art, because i suck at drawing despite having good taste, and as much as you improve with your hands i can improve with my eyes trough AI generations.
I think understanding styles and inconsistencies in your art is one of the central and most important aspects of creating decent game art, often when I see people say that they can't draw, their problem isn't necessarily that things are drawn "poorly" but that they lack an understanding of what they are doing. As I said, a beginner artist can probably use AI quite well, but you need to have developed some artistic intuition. I even make this point in the video, you mentioned "fashion" I mentioned director and interior designer, all are artistic skills that aren't as reliant on the craft, but they are still reliant on 'style' and 'meaning' which are equally if not more important aspects of art. I think most people have the insight to see that they don't like something, but that is a quite different skill from understanding what it is that they don't like about it. I'm basically just saying you need to develop the skill to recognise specifically what is 'wrong' with your art, which is a central skill to being an artist.
It's in the name: Artificial. Intelligence. It wasn't AI that came up with, "Garbage In, Garbage Out". I'd teach my own art to the AI, and use that personal assistant to help me get my own work done faster, if and when faster was something I needed. Close the windows, lock the doors, Go Go Robo-Tablet!
You can create pretty amazing assets in Midjourney, but as always, you get fantastic stuff quickly. And the more particular you are, the harder it becomes. But with inpainting you can also generate variations easily, ONCE you have the base. This could be extremely useful as you can get endless variations of say, trees or just tree-crowns which perfectly fit the tree trunk etc.
What I don't get is why people think AI art is any better or different than using pre-made assets. Except with assets it's clear legally, they look better, and you don't get throttled looking for more. You have to pay for the good stuff, it's never quite exactly what you want, hard to get matching art styles, it looks derivative, and everyone can tell. If you can get it to give you grass tiles that means they trained it on tons of grass tiles off the internet somewhere. You can probably just find those grass tiles and they'll be better. If the AI was trained ethically they'll be free or cheap to use. If they used copyrighted material without permission, you could just do that too if you don't care about the ethics of it. If you're using it as a starting point for concepts and doing the final piece yourself that's what actual references are for. And they're better because anatomy and perspective are correct. Every time I've tried to use it for anything it was just a big waste of time and I find that I work better and faster doing things the way I've been doing for years. I don't see any way it could improve enough to make it faster for me to make something I'm actually satisfied with unless it can literally read my mind. I think the hype around it is leading new artists to discover the use of references which in itself speeds things up and makes your art better than doing it from scratch from your head. Then they think the AI is saving them time when they could just have looked at some stuff on pinterest for free. (although now pinterest is filling up with AI garbage but hey at least you don't have to pay for midjourney)
Ye, I kind of agree. And I also kind of feel the same. there are a considerable amount of assets where the use of AI is just a waste of time. But I think there are those assets out there that take 10-20 hours to make, where I think artists might get forced to use AI for the sake of efficiency, either because their company forces them to, or their livelihood does (if they are self-employed). I don't think this is happening quite yet, but it feels like it is starting to.
Can we stop calling it "AI art"? Typing words into a blackbox and getting a computer generated mix of other peoples work is not art. Dumber than "ai artist" is only the word "promp engineer" which contains the word engineer to give the impression that it's a skill worth a six figure salary. I tried using generative AI for some concept ideas. It frustrates the shit out of me. Yes, the results can be pretty, but not what i wanted. If i need a central pillar in a cave with a ceiling height of roughly 50 meters, i want roughly 50 meters. Not 5 meters, not 5000 meters , not a random height each time i change the prompt, and not a glaring sun and clouds instead of a ceiling.
I mean I think we can maintain our beliefs and our values while still being honest about what is happening. I don't intend to do AI art, I don't really like it. but I still feel like it is worthwhile to assess and evaluate it.
I think this topic is a bit of a challenging one to handle. If AI is useful enough at accomplishing a task, it will eventually remove the need for humans to do that task. This has been the case with all technological advancements. Because of this I feel that it is kind of useless to grandstand. But I have wanted to talk a bit about the pitfalls with it and why I think it's both kind of "overrated" but also kind of not. mind you I made these assets in probably less than an hour, so I was actually kind of surprised at how quick it was to use. If you feel that it is obvious that this was AI art, and that it's terrible, I think realistically I could probably create a scene where you couldn't tell. I'm considering making a video where I use AI to create the best game art that I could, just to see how it stacks up, but I also have a lot of other topics I want to cover, so we'll see. As always, feedback is appreciated. Thanks for watching
I think you handled the topic very well!
I think that treating AI/generative models/machine learning like the boogeyman ultimately does a disservice to artists because true empowerment will come from knowing how to deal with this tech, rather than leaving it to ignorant profiteers and greedy corporations to do what they will with it. Of course the ethical concerns of how this tech is being implemented and used are important, but they aren't the only valid point of discussion.
Treating it like a serious threat also runs the risk of "validating" the idea that AI "art" is on par with human creativity, when the reality is that these models have no real understanding of what they're spitting out, and this will always be a limitation, since art isn't just aesthetics - as you pointed out, the artist's intention communicated through their work is just as important, so is the creative process itself.
The biggest problem with IA tech is that it only works if the majority of people don't use it, let me explain, what IA does is essencialy turn images into digital noise and then rebuild it, it does that with a billion images and boom you get the product, but you might have notice a couple of problems if the image especialy when there's repeting patterns, the IA kind gets weird with them, and not only that, but a lot of meaningless stuff just pops out, like the clouds on you're example, now if you imagine IA using IA images to train itself, wich is gonna happen if enought people start using it, especially if it's used in games and so on, them those little digital anomalies will start to happen more and more, this happens with any program if you make it interact with itself enought times, I'm a physics student and I worked with research for a period, this is a problem that happens a lot in physics research we always have to take a lot of care so programs don't start generating crazy results, it's acctualy realy easy to make the computer go crazy by making his output into input, in that sense is like inbreed children the problems with the DNA start getting bigger and bigger until the whole thing falls apart, and take into consideration that happens in scientific reserch the top of computer programing and data control, so IA's like DALL-E or midjourny are gonna start to become completely uselles with they become the main source of art, imo what's gonna happen is that the outputs are gonna start to get weirder until they become bassically just noise and the program can't turn it into anything, but it can also explode or smth, we really can't tell with this many variables, but if it starts getting it's output and reusing it as input the program is gonna fail, that' just the reality with computers, even if we think about them as perfect calculators in reality they can only count in 1 and 0s everything else is built with ducktape and prayers, and if you need anything with a lot of precision you're gonna need a LOT of prayer
I feel this is a fair assessment. AI always felt to me like the false promises of a "silver bullet" to not having art skills and being able to avoid having to actually put in the work. The continued devaluation of artists and the value of art was always there, it's just gotten more depressing nowadays with how obvious how much care folks put into quality.
Everyone familiar with the techniques have since the start - almost two years ago - been clear that the tools are best used by artists who also know how to make the most of them. It was never meant as a silver bullet. I think that is people mistaking it. Skilled creators can only rise higher with better tools.
I just hate it when I'm trying to do a search for some real life references and all im getting from google nowadays is some AI generated garbage
I've been feeling like google is terrible at references for quite some time. I mostly do nature assets and stuff so I tend to go to botanical gardens and similar and just take my own references.
Thank you for using the correct graph to describe the Dunning-Kruger effect! It's equally funny and infuriating how many people very confidently explain it completely incorrectly with made-up graphs.
Great video, I'm a professional animator in movies and I've had people ask me about ai potentially threatening my job as a cg animator, but even if they could get it up and running, they will lack the polish and the eye that comes from doing it for 10 years. With this being the case, do you think it's better for someone to ignore ai illustration completely and learn to develop skill and taste through hard work and intentional practice? Or would it be better to resume with ai and develop that "taste" you likened to interior design? And if you prefer the latter, how does one develop that eye and taste?
I mean as someone who likes to draw and will continue doing so I feel like it's tricky developing taste without actually understanding the craft at all. I suspect that the best course is still to learn how to draw and that that will stay true for a considerable amount of time. I'm just not entirely sure it is up to us, it is up to the consumers and the companies. Knowing how to draw will continue to be valuable, but I expect there will be pressure to use AI in order to maximise output.
@@Nonsensical2D Thanks a bunch for your answer, I want to learn to draw but there's a sense of futility with the impending ai programs getting better and better. Good to know there's artistic cultivation that mostly happens from doing, and even with a pressure to use it, drawing will still bring its own fulfillment.
@@MischiefMaker-mk9wn If your desire to learn to draw is driven by the potential money you plan to make with it. It probably won't motivate you long enough to get good. If you actually enjoy the process of drawing, feel the internal reward of creating something, of manifesting your imagination to the outside world - then it wouldn't matter how fast a program can generate pixels that it doesn't understand
Great video. The problem is that when artstyles of assets don't match, the game doesn't look good.
These algorithms are not precise by nature, at least not the ones there are right now. So it's super difficult to let it create a cohesive set of assets. It's not a matter of "just train it longer on more data", it is just not in the nature of how these algorithms work.
I appreciate the use of the correct Dunning-Kruger Graph
Haha, I considered using the "other" one, the incorrect one, because it is slightly better at illustrating the effect, but I knew there would be people out there (like me) that would be annoyed by that graph.
Thank you for making this. It's a touchy subject and I've long wanted to hear from an actual artist who doesn't just immediately put the blinders on and scream about AI stealing your lunch money!
I'm bad at art and frankly I'm ok with that. I'm a programmer with a ton of hobbies and I'm ok with not having this skill (or being musical). If I could snap my finger and be a great artist and musician I would obviously do it, but I can't. One of my hobbies is games (video and board) and having the ability to use AI to generate music and art is a great thing for me. All the stuff I do is for me and a few friends, so I'm not making any money on this. Hiring an artist or musician to generate this content (and it is a LOT of content that would takes dozens of hours of a human's time) is not something I could ever afford to do. AI is not perfect but it makes the things in my head possible when they wouldn't be otherwise. Two styles I often use is photo-realistic and plastic-miniature, which is not something most artists would do anyway (I don't think). Even using AI, I've still spent well over a hundred hours tweaking prompts and manually editing the results to get closer to what I'm looking for.
For what it's worth, I don't think AI will replace actual artists. It's too hard to get consistency. AI can be used to quickly generate "themes" or what not, to allow a game designer the ability to clearly express their vision, and then pass that off to artists to bring it to life. Same with music. But you'd need a human to keep everything consistent and put in the finer details and love.
My biggest fear is that AI will lead to even more overcrowding of media. We already have more shows/movies/music/books/games than anyone can possibly consume in a lifetime. Indie devs pushing games to Steam mostly just get lost in the flood. AI will make this 1000 times worse, with AI garbage art/music/programmed content being slapped together on what can just barely be called a finished product and then shoved out the door by opportunists looking to make a quick buck.
For what it's worth, I always thought the future of AI was to do the boring repetitive tasks and free up humans to pursue creative works, so this AI art/music/writing thing completely took me by surprise and I'm still kind of trying to wrap my head around it all. If AI is doing both the boring stuff and the creative stuff, what do we humans do with our time? 🙃
it's also worth noting that in their current state, generative models like "AI" image generators and LLMs are hugely inefficient, with every single prompt generating a disproportionate amount of carbon emissions compared to any other programming tool. it's an engorged version of your phone's text autocomplete suggestions, and unless you're curating a specific model from the ground up yourself, it's fed off of the labor of other artists without their consent or compensation. generative models definitely have uses, they're a powerful piece of software and have many practical applications , but they require stringent human moderation, are overhyped for what they are actually able to provide, have an incredibly destructive environmental effect when used at this scale, and honestly shouldn't really be called "artificial intelligence" in the first place. they're an early form of an imperfect, but potentially useful, set of tools that is currently being used excessively and poorly.
I had a similar problem with AI...I could not get it to understand "Isometric" Every attempt had perspective. I like your channel great list of videos👍
Can you show us your drawing process start to finish like making the canvas on procreate so we can see the size and then how large you make ur asset and what brushes you use I struggle with sizing without implicit references
Yes I am planning on making something along these lines soon.
The video really hits the nail on the head - I've been trying to work with AI to see how it can help me with my game art and have found it pretty useless so far, which definitely makes me feel more secure as an artist. The only real use for AI I've found so far is to create targeted reference images for my hand-drawn art. But I also agree that, while as AI develops it will be able to produce better art, there will always be a place for people who understand art techniques and have an eye for these things, or as you call it, taste. I've always thought that certain people are natural artists because they're able to see things in a way others cannot, and (we can hope at least that) AI is not capable of this. Or perhaps so far AI has mostly been trained by non-artists who also don't have that kind of eye?
We feel validated and acknowledged.
( I'm a 2D game artist myself. )
You described the unexplainable pain of artists so precisely.
I had a war inside me about whether I should keep making art as a career.
Your last sentences give me new hope. Thanks!
I never plan on using AI since it can't give me the unique things I want. I tried getting AI once to make a picture of a brontosaurus with gorilla arms and a giant eyeball for a head and....yeah....it gave me a vague dinosaur with like 3 1/2 eyes, etc. It was not at all what I wanted. So, if you want unique assets then humans still need to be involved.
I've heard people say it's good for programming so I tried to get it to make a spawn script on a timer and it screwed up a mere 4 lines of code. I still had to figure out the solution myself even though it did put me on the right path and I eventually got what I wanted. But that's leagues away from, "AI will make entire MMO's in minutes" claims.
Cool video, I am scared of AI as hell. I don't think that a non artist can produce cohesive scenes efficiently enough. But with enough research I do believe that artists can actually use AI as a tool efficiently. Combination of drawing for AI to well define the artstyle with more experienced prompting I do believe that it is possible, and that is exactly what scares me most. Artists would eventually become pressured by the time requirements and will start using AI, and that could start the downfall for all of us.
I am really hoping that AI usage is somehow legally managed, but I am afraid that train has left the station long time ago.
Ye, I wholeheartedly agree. I do like the fact that it seems as if AI can't own a copyright though.
I think the best uses for AI for artists is actually to make procedural repetitive stuff, then using that stuff to make your original art better. (Stuff like generating rocks texture for a 3d material, or to use with blend modes on photoshop, generating simple seamless patterns) Because some procedural stuff actually looks more natural when randomly generated than doing it manually.
But when it's something that you know what you want, it's super frustrating to try to micromanage the AI to make what you envisioned, so it's far better and more fulfilling to make it yourself. (And sometimes it's even faster than the whole prompting process)
That being said, I think you should try out Krea, it's pretty cool for testing out compositions and lighting. You can draw just the silhouette and see how it might turn out when it's fully rendered.
So I am attempting to do a 2.5D too down game. A cozy style game. I want to draw everything. How would you go about drawing buildings for the town. Thanks.
I subscribed because you actually teach. You are wholesome and honest with everything. Thanks!
There's this relatively older technique which is style transfer. A few years ago it was getting to the point where you could shade a ball and it would adjust a scene to the style in real time. I feel like all the full-on generative stuff kind took the whole spotlight, but style transfer is something much more interesting in my opinion.
Is there a course that you recommend (or is well known in the art world) that will teach you the fundamentals of good art design?
its a bit tricky. I talk a bit about it in a video of mine called how to learn game art, there are some courses like Drawabox, but I personally think just practicing and analysing gets you really far, especially if you want to do it for game art, and not general art. There are a lot of skills that are significantly more important for game art than they are for traditional art, so I think practicing normal fundamentals while helpful might not be optimal (and probably less fun).
Hi! Do you have any videos on how do you downscale your high resolution assets to 128×128 without loosing definition in-game?
A criticism of text prompts as an interface comes to mind, which is that it makes visual art(which is driven by our senses) into a social construction(language). We do this naturally in the process of drawing, to break things down and simplify them, but that process puts the artist "in the driver's seat" as this video aptly describes - making new language from the combination of the symbolic understanding and our senses. Prompting omits this basic connection. I believe interfaces that act to extend or stylize sketched drawings fare a little bit better, while still losing the information communicated when the artist is doing the rendering.
Do you feel like creating the assets in grayscale then doing your own thing with them would save much time ? I don't really like AI art in general and the moral implications but I have thought about using it as base art if it could speed up my game art creation some.
I'm honestly not sure, but probably. At least for assets that would take a lot of time to make. Cause you generally have some assets that take 3 hours to make and reducing time on those assets can be a huge time saver. For other assets like tiles, I personally feel like AI is quite useless, the prompting becomes annoying and it takes me 20 minutes to make it exactly how I want it anyway
hei im new to game desind can i ask what you use to desind your assets and what you use to put then together and making a game map and also i watch your video where you make runing animation using key frame what you use to put all the keyframe together.
you also but your assets, make maps and then add the running animation into it can you tell me how or please make a video about it thankyou verymuch
would be really helpful a video where you show your process from creating an asset to importing it into Godot. And also how you use tilemaps in Godot to structure your level with hand drawn arts. Thanks a lot!!
This is very much my experience. I've had some luck with getting quick and easy placeholder art for some projects where I was only interested in the code part. But it had exactly the same problems that you ran into. If you have a slumbering art interest it's honestly way more rewarding to create your own art and music. Aseprite is top of the line for pixel art and really cheap, and a midi keyboard and a free DAW like Waveform from Tracktion + some free instruments offers you music creation capabilities recording studios could only dream of back in the day.
I used to think AI art is also art but i've done a 180 on that take. Art is deliberate, there's thought and meaning behind every minute detail. There's no art without an artist and AI is not an artist, it's just that, an algorithm. I hope that AI doesn't improve further that CEO's of AAA studios lay of their artists in favor of AI to chase more profit. It would suck ass
I think the question whether 'AI art is art' doesn't make much sense. As you said, art is deliberate. AI is a tool, and if you play with it enough, you'll know that lazily prompting doesn't give you the visualisation of the idea you want to portray. The art-iness comes from the person, and when people say the images look bad, it comes from the person being bad at it. Asking whether 'AI art is art' is like asking whether ink being splashed on a wall is art. It might or might not be. If you just bump into something and spill the ink, that's probably not art. Trying to define that all AI generated images either is or isn't art is misleading.
And regarding CEOs laying off artists, I don't think it's about the improvement of AI. That's the vision (or near-sightedness) of the CEOs. Generative AI is a very powerful tool for human creators, and it should be improved to the fullest of its potential, while still be ethical.
@@kieselzusammen This is a quote from a sculptor I saw in a video once "It can feel a lot like magic, because when I move my hand, I place my chisel and when I swing the mallet I know exactly how the stone is going to break. It doesn't make any sense, but I do. I visualize it, and then it comes true."
AI algorithms, among other things, are dependent on data sets. They're not made out of anything as part of some natural process, that follows natural rules. They're not stone, they're not even ink. Tools are instruments wielded with precision, used for craftsmanship. Imagine you had an algorithm that could remix cooking recipes. To me calling the conveyor belt of data that remixes art "tools" is like calling the recipe remixer "ingredients."
As for CEOs "ethics" is not in their job description.
@@futurestoryteller Tools are anything that make your work easier, including automation. AI can also make precise images, even without you drawing over parts of them yourself, but you have to be very precise on how to prompt and use other extensions to do it. So it's just not that good a tool yet, but if we don't count this a tool, then editing photos with some Photoshop tools like Content-Aware Fill doesn't count either.
The remixing of data is also kind of inaccurate. Many AI models work by denoising randomised pixels until it fits the patterns it has learned, but it's not exactly just mishmashing the original works. If we think about how we know what the correct highlighting and shadowing should look, we don't work that much differently. We saw a lot of patterns of how light and shadow should look, and we put it into image in the way that should make sense in the context of the thing we want to portray.
The difference is AI doesn't understand context. It doesn't know what makes sense, but it isn't necessary for it to do. Being a tool means it's already okay if we let all the context-related considerations be the user's responsibility. Actually, that's how things should be. You shouldn't go 'look ma, no hands' when you try to create something.
The main reason I disagree with the notion that it just remixes art is how it's impossible to store the amount of training data those models use within the size that they really are of. You can't combine two things together if you can't access both things. Those AI models can't reproduce the original works they are trained from. (Technically, they can, but only when they are 'overtrained' which means something is just too prioritised that they ignore other training data, so the vast amount of original works still couldn't be reproduced.) Everything is just reduced to parameters and equations representing what it has been trained on.
@@kieselzusammen A person who does all my work for me would make my job incredibly easy. If I've done that, am I the real artist, and the person the tool?
Saying we simply recognize patterns and copy them is a reductive take on human consciousness, if not the skills necessary to make art. Computer graphics software can simulate shadows and reflections better than humans, the only reason we don't pretend they "understand" the data we've plugged into them is because they're not marketed under the pretense of simulating actual neural networks.
They are not at the user's discretion, as demonstrated by this video, the fact that it will never actually understand what you want, while it relies exclusively on what you tell it (as well as what data it's collected) means you are always at its mercy, not the other way around.
Those parameters and equations are the data it's remixing. That reduction of size is a red herring. I'm still pretty sure you're taking something that's a more complex version of something completely normal for computers, and because of the complexity with which it operates, kneejerk ascribing virtually anthropomorphic properties to it. Image and video files are compressed all the time. But because this time it's complicated enough to _appear_ abstract we're just accepting the facsimile as "authentic" to something other than normal compression. A good example of AI proving this concept in practice is "deblur" and "depixelate" AIs. If you have blank film frame a normal compression algorithm notes a pattern and reduces the number of pixels it deems significant to reconstruct the overall image, right? AI is doing that, but for more complex patterns of pixels.
The fact is it's just doing what computers have always done in a fancier way. It doesn't see or understand the outputs. It's still just made up of numbers. It's not an art generator, it's an art calculator.
@@futurestoryteller My point wasn't 'people also copy patterns,' but 'AI doesn't just copy patterns.' It is clear you think AI does that, so when I compared it to people, you thought I meant people just do that, which is the opposite of my point.
> They are not at the user's discretion, as demonstrated by this video, the fact that it will never actually understand what you want, while it relies exclusively on what you tell it (as well as what data it's collected) means you are always at its mercy, not the other way around.
I'm not sure how you think AI as a tool is supposed to work. Should it be self-sufficient? Then it wouldn't be a tool anymore. Other tools 'rely exclusively' on our control as well. Currently the most common way to control the AI is just through language, but it isn't the only way you could do it either. The example of the tree in the video might be able to be done if you can use some extensions like inpainting.
> A good example of AI proving this concept in practice is "deblur" and "depixelate" AIs. If you have blank film frame a normal compression algorithm notes a pattern and reduces the number of pixels it deems significant to reconstruct the overall image, right? AI is doing that, but for more complex patterns of pixels.
Except that isn't how it works, and I doubt if they do that, we'll get the same results. First, suppose we want to reduce the size of the original works, we can do it lossy or lossless way (meaning if we lose any quality in the data we store.) Lossless is out of the picture. We don't have an algorithm that could reduce the size that much yet. So if that's how we want to do it, we'd have to store the images as the lossy ones. And the problem with reconstructing the lossy images into crisp-looking ones is how, as we know, machines are dumb. If we just let it guess how a blurry image should originally look, there's almost no way they can get it right, especially with details like human face. It needs to know how high resolution images look so it can remake the low-res ones somewhat convincingly. Now the problem is recreating itself, because if it only stores reduced quality images, there's nothing for it to compare with so it could reconstruct how the image should originally look.
I'm not under the impression that it's sophisticate enough I anthropomorphise it and say it does things like humans do, but your speculation on how it works couldn't be correct due to inherent technical limitations.
And if we get abstract enough to say that even if it finds out patterns and store mathematical parameters, it technically is just compression, then this is so broad that any kind of learning, even human learning, could also fall into this category. Because we still have to store what we learn physically. Researchers can even reproduce images from our memories, even though in very specific conditions and with poor clarity, so we also 'compress and store' what we learn.
Again, the point isn't 'it works like humans' or 'humans just remix from memories too.' The point is 'what it does isn't just remixing what it stores,' because that doesn't work.
> It's not an art generator, it's an art calculator.
I completely agree with this.
I tried to use AI for creating assets like rocks, trees. However the style and perspective does not feet my expectation.
As you said, it is promising, but from my point of view it is still not ready for helping.
The one way I see it can be used now is just for creating some reference, or creating drafts.
Stacked with creating background and levels. Before cooperating with a human artist, I tried to use AI to understand how it should looks. No results =(
The only time I find AI art anything close to useful, is just for a moodboard or first draft concept art.
Like for instance, maybe I meed a monster. I prompt midjourney to give me an image of a monster. Then draw my own version of that monster with better proportions and details that are easier to read, then animate it.
On one hand, it can potentially get rid of the creative block at times.
On the other hand, I tend to get my ideas easily without having to spend hours on any of those tools.😅
You struck a chord with me. I have no idea how to identify identical artwork, I just know some things are off when I see it. Now I wonder if buying an asset makes anysense at all.
That trick of having the AI generate greyscale mostly except maybe if you are in the idea phase, then coloring in engine is pretty great.
With the example you gave for AI art the worst part is probably the ground for the platform. But if you are going somewhat high res but without too much detail anyways. You could always just lay down some paint brushy gradients and use that. And then use the AI for anything that takes a lot of detail.
the example you gave of simple hand-made art looks really charming though. And for the right game would look way better while being fairly easy to do.
I completely agree it is pretty much unfeasible for the character stuff.
Although cartoony character ontop of a more painterly or detailed background is a classic combo that works, so if I was going for AI on a game. (which I am not.)
Doing something like that would be sensible.
Overall, I feel like painting skills matter a lot more now that AI is a thing.
Drawing also matters but basically only fundamentals like construction and not much more.
Being able to block out rough colors/areas is valuable.
Since in the context of a game, the assets themselves have to be used in context anyways. And having someone think about the role each asset plays is pretty important for making the assets work together
Most likely, to create actual game art, you'd probably want to use more advanced Stable Diffusion workflows where you can actually create finetunes or embeddings to force everything you generate to have an unified, common artstyle based on a selection of examples, either generative or real.
AI usually comes with a nice expressive palette. The art presented as human made on the other had has this low contrast moldy fungal feel. But it somehow got popularity with the games like Cataclysmo. Apparently indie artists now go for shock value, in both form and substrance, instead of pandering the spoiled consumers with nice aesthetic pieces. Then again, AI was explicitly trained not to produce revolting palettes
Amazing. This is what I’ve been thinking. This is what I see with people who use ai. They don’t know what looks good at all.
This is actually the exact reason I've given from the beginning for why AI can't make art, in any literal sense, because an artist has discretion and a neural network doesn't understand the meaning of the word. It can spit out seemingly infinite variation on the definitions that it's culled from the internet, but it doesn't actually _understand_ it.
It's like if NASA trained a computer model on information from every space mission. It would be able to dispense fixes for common problems, or fix bigger problems that NASA has probably already implemented solutions for, but it wouldn't be able to understand a truly unique problem, something outside its usual data set. It's not thinking like a human
do we have a discord channel yet?
The web based interfaces such as Dalle are essentially just toys for one off images and useless for this sort of productive work, sort of like RPG Maker isn't what you'd use for building a first person 3D game. You'd want to get specific models and likely finetune them to reliably produce the style that you want, or use the various guidance techniques such as ControlNets. You'd also generally spend a lot of time either taking it over to Photoshop, Affinity, etc, to do touchups or change the direction of things before bringing it back over, it's very much a back and forth process. Some interfaces like Invoke seem like they'd be better for working in place without switching to other tools. You'd also want to be careful about which model you pick based on the image compressor/decompressor it uses, e.g. the 4 channel VAE of Stable Diffusion 1, 2, and XL can't do flat shaded artwork very well, but the newer 16 channel VAEs are great at it.
Very interesting video, it kinda put into clear words a feeling I had for a while with AI generated content. As an artist myself, I'm fairly interested in the matter, mostly from a distance, but still, interested.
If I were to make a video game, which may or may not happen someday, I feel like using AI to generate bases for some assets or some variations of existing design then repainting on top of it, to clean it, add that "intent" and such, might be a decent option, especially if tackling a project like a video game on my own. Not sure I will, but, good to know it's possible and doesn't look too bad.
Well, yes, AI is no more than a tool. A REALLY multimodal tool, but a tool nonetheless, you need to have skill in the area that you are using it for to guide it correctly.
I use AI to aid me in programing, and it solves a lot of things, I've learned a lot thanks to it too, but if I did not know what I was doing I could not obtain anything coherent from it.
You don't even need to be good at art, just to have a basic understanding of composition.
I also recommend against it because it's a swamp of copyright issues.
Why dont you try this again with stable diffusion and some other good Lora models?
When I check collaboration forms, I see more "for hire" posts for programmers, and more " I need 3d-2d artists "
Im not a good artist. But i can take a photo or ai image and make my own art based on it.
But i agree that just directly using ai assets could be difficult
My biggest problem with ai art is how artificial it looks. Sure it's very detailed and looks professional, but it also looks really dead. I honestly would much rather see someone be bad at making art than to see a "well drawn" ai piece of art, because the bad art of something like an anime character or a cartoon character at least has heart behind it. A machine is not alive and only can make art through the prompts it receives; therefore, ai art just looks souless in comparison.
It doesn't mean I'm threatened by ai art or I hate people that do ai art for fun: far from it. I just prefer seeing someone better their skills, rather than using the easy button.
it just copies the stuff the companies stole from artists. It has no intelligence, it doesn't think, its just a probability algorithm, basically it auto completes your sentence with a picture or in the language model with text. AI imo stands for artificially inflated. It will be used as a replacement for stock images, or to create mood boards when the bubble bursts. I already see companies in the game industry that struggle to finish their projects because they thought they could replace their artists with ai prompters, now they rehired artists to fix the bullshit that the ai creates but somehow still insist on having ai prompters because saying you use ai gets investor money...
Yeah there were a few weeks where everyone was enamored with AI art online. But after that it was like eating cake for every meal. People got tired of it and the groups I am part of started banning it.
@@Tinyflower1 Though stealing is okay when it's intellectual property from corporations, right?
That guy is scummy trying to force ai to do your art style. They should just do their own art instead of stealing from others.
In my opinion AI is a good tool to support artists, but a terrible tool to replace them. AI can generate little environment things and other side stuff, but AI messes up, A LOT. Having an artist to detect and fix up those flaws is necessary if you don't want every aspect of your visuals to have some thing that just looks... weird. Otherwise you end up like Once Human where, if you take a closer look, assets are SOOOO missmatched that it's honestly baffling to think about.
It's refreshing to see an artist have a realistic take on AI art without instantly dismissing and moralizing it. Overall, I agree with your main takeaway, although you might have been a bit limited by only using Dall-E. The process of simply generating an image using only the prompt and copy-pasting it into the game is deeply flawed, due to stylistic inconsistencies (that are going to be instantly noticed by the player, especially if it's a foreground sprite).
Perhaps a more controlled approach could work, by using a local Stable Diffusion model and processing a rough handmade sketch through a pipeline i.e. sketch -> ControlNet/img-to-img -> inpainting-> background removal -> manual editing. I also think that using a specific model or model finetuning (e.g. Lora) to achieve a specific style might help in providing consistency.
Maybe using AI art not to create base art, but variations of existing sprites could also be a better use case. For example, different customization options for the character or types of vegetation for different biomes. Something I would really like is AI to be able to create different poses and compositions of input character sprites (for animations too), but from what I have seen and tested that does not seem to be possible currently.
As it stands, AI image generation is a tool that is greatly overrated by some people and underappreciated by others. But I don't think it is capable of surpassing us in the near future and completely replace us. And I sure as hell will not let AI take away my passion for gamedev and drawing, regardless of whether I make money with it or not. If it comes to that, I'd rather become an advocate for economic reform (such as introducing UBI), than to demonize and shun everyone who makes use of technological advancements.
Would you be willing to retry this experiment with Stable Diffusion's advanced tools? A lot of the issues you faced might be fixed using features that Dall-e doesn't have (eg. using inpainting to fix pieces that would be great if not for a single detail, use scribble controlnets to construct the basic structure of the asset before generating, use seeds to create slight variations of assets, or even train a lora based off your own art to train the AI to follow your style, which is something you said you were interested in).
I agree with your statement about how much composition is important to art, which is why I'm especially interested how an actual artist with a trained sense of style can use the tool in a production pipeline.
Interesting. Ye, I could look into it, I'm a bit unsure how well it could run on my computer though (I only have 4gb). But I also think you could get significantly better results with dall-e than what I produced here, if you spend some time with it.
@@Nonsensical2D One of the big issues I found with Dall-e is that your access to it is filtered through Chat-GPT, meaning you type your request to Chat-GPT, Chat-GPT creates a prompt based off your request ("Make a game asset of a bush with a simple artstyle" becomes "game_asset, bush, simple_artstyle" or something), and then that prompt is shipped to Dall-e. It makes it really hard to experiment with since you're not 100% sure what it is actually being requested at any given time, and your response may or may not be influenced by your previous requests.
A lot of maybe for the conclusion.
The future is a lot of maybe
I've been telling people who say 'AI art is crap' that it's not about AI, it's about the knowledge and attention to detail of the person behind the AI. If you think the images look bad, it's often because the person doesn't know why it's bad or how to fix it, rather than because AI can't be used to make decent looking images. But my voice lacks weight because I myself don't know art very well. Glad to see your take is rather close to mine.
Anyway, even if you lack the skill to draw over the bad stuff from the AI yourself, you can still draw over it badly and then make the AI use it as a reference to fix what you drew. I'm not sure if Dall-E can do this, but Stable Diffusion can.
its similar to when someone just takes a photo and traces it or comor picks. without art skills it looks really bad. same thing with rotoscoped animation without knowledge of 12 principles of animation - it always look bad. same thing with just copying code from stackoverflow/ chat gpt. you need to know what you are doing even if ure just supervising it
AI images look really impressive until you've made your first asset by yourself. No matter how crummy it might look, it'll be and feel 100 times better.
wait for yhe improve of the ia or draw it your self, is more satisfaying
Wow. This is exactly what I felt when using these prompt made art AI programs. I’m a game dev and also draw art, but even if the AI art is “good” sometimes it feels off. So in the end you need your own skills to actually use it in your game, if you want to do that.
Never late to learn new things I guess. Thanks for the eye opening video. It was great to know somebody actually having the same thoughts.
I guess the road of art and game art is a tough one. But we should never give up keep trying to get better at art!😊
I agree that AI is inevitably here to stay. The worst part of AI for arts (graphical, musical, written or any creative art) right now is how it is trained by stealing the work of other artists without consent and giving no compensation for their work. All that said, AI is a tool and if they correct the legal and moral part of training it with other artists art, it can became a very usefull tool. I totally agree on that I prefer art created by humans with intention and meaning and I am also totally bias because I enjoy drawing. It is the process of creating art what is the most important, more even than the result itself. I also sadly agree on creative skillful artists will have to adapt to this new AI world. But at the same time, I think, and hope it will be a growing number of people that not only prefer but only want human created art. And I tottally agree AI is not a golden ticket on any field. If it is not guided by an skillful person that understands the specific task at hand it become mostly useless.Thanks for the video.
Ye, I actually somewhat speculate that artists that create paintings (like oil, acrylics, watercolour) might be somewhat unaffected for quite some time. At least when I've sold personally, 500+€ are sums where the person buying often want your signature and your craft to show through, otherwise they would have bought a poster. (though that side of art has always felt like a tricky to make a living).
The more we look into it, the more we realize AI isn't really intelligent at all. Calling it "AI" is a marketing ploy and a misrepresentation of what it really is.
Ye, for sure, I think I make a point to use the term generative AI (or LLM). I think at this point when discussing intelligent AI we just have to default to using the term AGI. It's kind of similar to the euphemism treadmill, but a kind of 'marketing treadmill'. Which is why you also see some people like Linus (LTT) use the term ANI (artificial narrow intelligence) to refer to OpenAI-stuff. It is what it is, but I agree.
@@Nonsensical2D yeah. Personally, I like to use the term "garbage" lol. Your attitude is more sensible and makes your point come across very well.
I like that you did not shoot AI down like most other artists do. Personally I use AI to help get ideas out of my head, as a starting point.
And when ai game dev becomes a thing you'll be absolutely OK with it yeah?
I mean I don't really like AI art, I don't necessarily understand what AI game dev actually means, but I'd probably not like that either. I'm significantly more of an artist than I am a game designer, so AI art is actually a bad advancement for me, but It's not like I complained when digital cameras or phone cameras became a thing, so it'd be kind of hypocritical to complain if technological advancement does successfully replace me, it's not a world I'd prefer, but it is what it is.
@@Nonsensical2D I get what you're saying but when digital cameras came into existance they didn't create photographs by stealing other peoples work... regardless how we all feel about how AI looks at this point I just can't get past the fact that it's stealing from people like you and me (I'm an artist as well and I've done a few games). It's not really about being an uptight artist, more that our work and labour is being stolen and exploited while we continue to be underpaid... it's a bit different than technological advancements like cameras because they weren't built by scraping, u know? I learnt a lot about assets from you and I hope we get to learn more about creation in the future from you and not AI. Obviously do what you'd like but idk its sad to see so many artists take this direction lately on youtube.
@@lesflaya I mean I agree with you, and I'm personally not planning on using AI, I like to draw, it's what I do. I frankly don't like this development, but I also think the battle is futile, the only people that care are the ones immediately affected, the artists. The rest of the world just want a good end product.
@@lesflaya While many current image generating AI models were built on scraping, I'm not sure if it's all of them, and the way they work doesn't inherently need scraping or unauthorised use of images as training data. Some people also already made checkpoints for Stable Diffusion using only authorised and public domain images. Those checkpoints simply aren't that popular yet compared to the questionable ones. Would you still feel the same way about the models that don't steal from anyone? Because I personally believe that sooner or later, they will become more mainstream and might be able to replace the models built on scraping.
@@Nonsensical2Dyou think oversupply, thus, tragedy of the commons wont impact usage of AI?
I have to disagree with your point, especially your point about the Dunning-Kruger effect. I don't have to be a musician to tell good/harmonious music from bad or music produced by an amateur or beginner. Same with art. I am no art critic, yet I can distinguish a 5 year olds drawing from that of a professional painter. Of course there may be an expert level where only experts and artists themselves can appreciate certain techniques or flourishes to the art. But that's not the main demographic of ... well, anything. I think A.I. art won't take over, it's not smart enough. However it will equalize and liberate people. Sure, bespoke and custom art will always be better and allow you to realise your vision with pinpoint accuracy - for a price. On the other hand even hobbyists will be able to create something with A.I., though the trade-off might be that it won't be as pretty and cohesive as with bespoke art.
I think the problem arises not when you have to distinguish a 5 year olds drawing, but being able to distinguish something good from something less good, or something good but which contains something that is "off". and also knowing how to handle it. Suppose you have something that looks nice, how do you know how much nicer u need to make it until it is 'just right', how do you know when to make something less nice in order to increase cohesion. This is a problem that might exist less if you use AI art to create an individual art piece, but when you are making a game you have a whole host of issues that you have to deal with, because you don't have one art piece, you have 500 assets that all need to go together
Your argument ignores a lot of nuance in what we see as the end user versus what went into creating that complete picture. Humans are very good at interpreting whether or not something is "off," but without training, we don't have the language to describe what's "off" about it. I think that was the point of the video.
Like Nonsensical2D's comment mentioned, you're not feeding high level information into an AI. At this point in the development of generative tools, you're having to describe, in detail, how you need something tweaked to get a good end result. If you're using generative AI for your game, you're essentially occupying the role of Art Director on your project, a role that requires decades of professional experience to communicate with the nuance required to achieve a good final result.
@@calicow And that's where I disagree. I don't need decades of experience to see if something looks bad/off and yes, not everybody will be able to create coherent and good looking art, even with support from a.i. or other tools. However there is a wide margin, even between professionals. I would guess A.I. will, in the coming years, replace low skilled up to medium skilled professionals while simultaneously liberating the market and opening it up for a lot of semi-professional developers using the technology to create something. This is similar to what self-publishing and ebooks did for the book market. Is there a lot of sub-par low quality stuff? Sure. But on the other hand there are also AAA games and traditionally published books I wouldn't play/read if there was nothing else to do. This is just another tool in the belt. I, for one, am no musician. So if I can just pump out a passable tune with the help of a.i. then that is more than I could accomplish myself and if it's something for a small scene, that might be all I need.
lol remmove
I think you should retry this experiment with Midjourney, I bet your opinions would be different
I disagree with the central point of your thesis, which is also speculative in nature. You dont really know that people who cant draw dont have an eye for a creative process. To make an example; I cant sew anything, but i know how to dress nicely and properly. Same apply for the overwhelming majority of people who know how to dress.
i'm extensively using AI and i can see when something is not working and doesnt fit the style of previously generated assets, despite the fact that i cannot draw. It seems to me that you missed the inconsistency of the point you were trying to make: you said that when people look at AI generated collage they can see it's not deliberate therefore less aesthetically pleasing. But who are this people? are they all artists? obviously not, everybody can discern good art from bad one, although it definately takes experience to discern from very good one to just good. The example you pointed out are just bad use of AI, mostly because it's an incredibly frustrating endevour, sometimes the AI dont listen to even basic instruction after 500 different iteration of the same thing, and so they might resort to give up and try the least bad attempt and mesh it together with the rest, but that's just the same as any other aspect that revolves around the creative process.
Many art critics couldnt do anything even remotely close to what they criticize, yet they have the eye to discern if it's good or not, if your point was valid that wouldnt happen.
I'v seen insanely good AI projects; there was a test done times ago where asked people to discern from AI images and real, and nobody could. there was a painting competition few months ago where an AI generated work won over "real" artist.
Personally AI is the only tool that i can use to generate art, because i suck at drawing despite having good taste, and as much as you improve with your hands i can improve with my eyes trough AI generations.
I think understanding styles and inconsistencies in your art is one of the central and most important aspects of creating decent game art, often when I see people say that they can't draw, their problem isn't necessarily that things are drawn "poorly" but that they lack an understanding of what they are doing. As I said, a beginner artist can probably use AI quite well, but you need to have developed some artistic intuition. I even make this point in the video, you mentioned "fashion" I mentioned director and interior designer, all are artistic skills that aren't as reliant on the craft, but they are still reliant on 'style' and 'meaning' which are equally if not more important aspects of art. I think most people have the insight to see that they don't like something, but that is a quite different skill from understanding what it is that they don't like about it. I'm basically just saying you need to develop the skill to recognise specifically what is 'wrong' with your art, which is a central skill to being an artist.
It's in the name: Artificial. Intelligence. It wasn't AI that came up with, "Garbage In, Garbage Out".
I'd teach my own art to the AI, and use that personal assistant to help me get my own work done faster, if and when faster was something I needed.
Close the windows, lock the doors, Go Go Robo-Tablet!
I have found AI art worse than useless. Its an ineffective time sink.
I think traing and prompting ai will be its own skillset
Great video!!! Thanks for your work!!!
You can create pretty amazing assets in Midjourney, but as always, you get fantastic stuff quickly. And the more particular you are, the harder it becomes. But with inpainting you can also generate variations easily, ONCE you have the base. This could be extremely useful as you can get endless variations of say, trees or just tree-crowns which perfectly fit the tree trunk etc.
Appreciate the absence of cynicism towards AI.
What I don't get is why people think AI art is any better or different than using pre-made assets. Except with assets it's clear legally, they look better, and you don't get throttled looking for more. You have to pay for the good stuff, it's never quite exactly what you want, hard to get matching art styles, it looks derivative, and everyone can tell. If you can get it to give you grass tiles that means they trained it on tons of grass tiles off the internet somewhere. You can probably just find those grass tiles and they'll be better. If the AI was trained ethically they'll be free or cheap to use. If they used copyrighted material without permission, you could just do that too if you don't care about the ethics of it. If you're using it as a starting point for concepts and doing the final piece yourself that's what actual references are for. And they're better because anatomy and perspective are correct.
Every time I've tried to use it for anything it was just a big waste of time and I find that I work better and faster doing things the way I've been doing for years. I don't see any way it could improve enough to make it faster for me to make something I'm actually satisfied with unless it can literally read my mind. I think the hype around it is leading new artists to discover the use of references which in itself speeds things up and makes your art better than doing it from scratch from your head. Then they think the AI is saving them time when they could just have looked at some stuff on pinterest for free. (although now pinterest is filling up with AI garbage but hey at least you don't have to pay for midjourney)
Ye, I kind of agree. And I also kind of feel the same. there are a considerable amount of assets where the use of AI is just a waste of time. But I think there are those assets out there that take 10-20 hours to make, where I think artists might get forced to use AI for the sake of efficiency, either because their company forces them to, or their livelihood does (if they are self-employed). I don't think this is happening quite yet, but it feels like it is starting to.
Can we stop calling it "AI art"? Typing words into a blackbox and getting a computer generated mix of other peoples work is not art.
Dumber than "ai artist" is only the word "promp engineer" which contains the word engineer to give the impression that it's a skill worth a six figure salary.
I tried using generative AI for some concept ideas. It frustrates the shit out of me. Yes, the results can be pretty, but not what i wanted. If i need a central pillar in a cave with a ceiling height of roughly 50 meters, i want roughly 50 meters. Not 5 meters, not 5000 meters , not a random height each time i change the prompt, and not a glaring sun and clouds instead of a ceiling.
Damn I guess it's over for this channel huh.
AI is just bad, there's nothing good about it.
I mean I think we can maintain our beliefs and our values while still being honest about what is happening. I don't intend to do AI art, I don't really like it. but I still feel like it is worthwhile to assess and evaluate it.
☹thumbs down and unsubbed.
Why? You hate calm people talking?