I just realized that there's an addictive element to generating AI art. The fact that you only occasionally get good good results gives the whole process a slot-machine like aspect. It's like an artistic loot box.
It's all the imagination of a human artist without any of the common sense. Honestly, I think "artificial _intelligence_ " is a misnomer, even the internal mechanisms are much closer to how _intuition_ works.
@@TubususCZ In my opinion these models are pure intuition without any higher cognition. The network structure does not allow any higher cognition, so this is not surprising. I have dones some tests and a GPT network gave remarkably similar results to a human, when the human is not allowed to think much. Basically we limit the human to only able to run a single pass through their brain, no deliberation. The networks usually do use multiply passes (diffusion models, use diffusion steps, GPT usually uses beam search, similar to a tree search), but these passes are for increasing the power of the forward pass, basically amplifying it, implementing higher cognition with them is extremely difficult (or maybe not even possible). So these passes are amplifying what is already there, not really adding an extra layer which would be necessary for more abstract thought.
@@thunderspark1536 They're not smart but they spit exactly what was fed into them. Lots of Stable Diffusion models show that if you train the AI to do something, it will do that thing incredibly well. If someone had a Stable Diffusion model trained on medieval weapons like swords, axes and whatnot, you could be certain the AI would be able to spit out high quality and consistently practical swords without much effort. The reason Shad is getting very random and not consistent results is because he's using a Generalist Model.
@@muatring Which is how children do things, as I said. If you have a child grow up to specialize in sword making they can make some very lit swords, while a grown up specializing in chemistry wouldn't be able to. Same story here.
Seeing those swords that didn't have handles made me think of something interesting I saw in an old fanfic once. One of the races in that story made extensive use of their natural telekinetic ability in combat, and designed some of their swords around that. Those blades actually lacked a hilt entirely, simply being bare blades sharpened their entire length and a point on each end. This prevented an opponent from simply overpowering their telekinetic grip and grabbing the blade to use it against them.
I can already imagine the perfect weapon for a telekinetic individual: a detached knife blade, about six inches long. Very deadly, even with the assumption that you can't accelerate it that fast, and almost impossible to defend against
I thing a great weapon for a telekinetic (depending of course on how the telekinesis works) would be something tiny, like a single grain of sand. Imagine what they could do just by indiscriminately wrenching it around within someone. I ninja star would be pretty great as well.
Yeah, I’d definitely want the villain I’m fighting to have to hold onto his own sword blade because there isn’t a hilt, or to hit me with the hilt because there’s one at either end, lol
Villain: why do you approach m- Blacksmith: SHUT UP! Your weapons are a *disgrace!* Villain: Wha- B: I brought my anvil. Just tell me we’re you DROP YOUR WEAPONS, you imbecile! Villain: [points]… do as you like, you can scold them if you want. B: oh, YOU are getting yours when I am done… XD
A long musket with a specialized stock designed to be used 1 handed. The musket is loaded with a roll of 40 coins. The coins are lightly enchanted to each home in on a single target like flying disks. A bayonet is fixed to the musket that is enchanted to shoot Lightning Bolt 75 times per fully charged. In the other hand an aspis shield. The front layer is steel behind that bronze behind that birch behind that pine behind that birch behind that 37 Giant Spider Silk sheets sewn together in the quilted pattern and glued on with 1 of such pads. Behind that a plug of leather armor and that is where the straps are. All this is just 1 aspis shield, however the entire thing is actually 3 aspis shields glued together. This 3 fold shield is enchanted with the spell Shield. When the spell Shield wears out the shield is also enchanted with 45 days worth of Mending Cantrip. It will just keep repairing itself each time it is hit for a while. As a backup weapon a Shad style Bastard Sword also enchanted to shoot Lightning Bolts 75 times. Needs a helmet. Suggestions appreciated.
I'm something of a fantasy villain myself, so thank you. 5:27 is perfect for my undead army. They don't really care if they hurt themselves more than the target.
Man, reigning in A.I. art is funny, sometimes you get good work and other times you get abominations comparable to the reaches of the human imagination. Makes me remember a comedy quote of "Playing god is a SOMETIMES thing."
Trying to get an image AI to produce something decent really does feel like trying to find the right incantation to summon something. You start to feel like a techno-warlock, drawing power from something you don't really understand.
My personal favorite is at 17:43. The cross guard could be wider and the pommel a little bigger, but I really like the way that the entire sword looks like it is all from one piece metal. It looks almost like the sword was grown as opposed to crafted. The light across the blade looks a lot like Valyrian steel, with ripples across the blade. No idea how practical it is, but I like the concept of the entire sword being fused together instead of distinct pieces added together.
The important thing to remember with an AI art program is that the AI has no context for what any item is or what it might be used for. It just has geometric trends based on image libraries in it's sample sets. So while it can make items with sword like properties, it really has no idea what a sword is. So it's pulling any image with the tag and throwing common threads together at random.
It doesn't actually have any images and tags available after it's deployed. It's just 3GB large and works while offline. As such it cannot contain any image library of meaningful size
@@PacMonster0 the point stands that as long as it's sample pool is strictly images, especially static images, the AI will never understand that a sword is "a tool intended for use by a human to assault that has a blade at one end, a handle at the other, and usually has a guard between the blade and handle." It doesn't have the context to draw those conclusions. It will just know, "these geometries have a high association with the node labeled 'sword'." Which will always lead to chances of unrestrained images like those above having a nonzero chance of being produced. Just like other fantasy art creations shown on this channel before, it simply didn't have enough consideration for what a sword is, what makes it effective, or how it is used. Just, moreso in the machine's case. Also, training on the amount of data required to avoid this is would be at a high risk of overtuning, which effectively renders the AI unable to take any liberties in its actions or predictions.
@@PacMonster0 Sir. I made the initial comment to which you replied. I think I have a fair idea of when the point was meant to be strictly images, which I apologize if I made this unclear with my first comment and failed to clarify with my second. As to what I understand, I just finished a course on Statistical Analysis and Artificial Intelligence. The majority of my second comment was a literal rundown of the 'thought' process used by neural networks. The sample data, training data, and application input steps require breaking concepts into mathematical algorithms and then comparing the outputs of these algorithms for the given inputs against each other. As a computer program, they have no understanding for any label that is not expressible as a number. Image geometry can be defined numerically with comparably little effort (though still not inconsiderable, Three blue one brown did a very good series on this. First video starting th-cam.com/video/aircAruvnKk/w-d-xo.html even my instructor recommended this). Combat functionality, the grip comfort for a fictional object as expressed to something without a tactile sense, that an object requires a grip at all, or the very concept of what a grip even is, are far more abstract and do not readily conform to numeric metrics. The fact stands that an AI, much like any other intellect, will never get to the point where it cannot make a mistaken inference when there is crucial data that does not exist within it's experience set. It can be very unlikely. It's by no means guaranteed that the mistake will happen. It may even have a near zero probability. But my point was that a mistake is still always possible. This was one of the most important points expressed in the class I took. AI are useful and quick at making accurate assessments. They are not perfect, and to treat any of them as infallible is a dangerous mindset. Apologies to Shadiversity. I've always held this channel up as one for preventing partial or poorly contextualized information leading to incorrect or inaccurate assumptions. My comments have clearly failed at this level of communication.
@@PacMonster0 Machine learning is at it's core a process of trial and error receiving feedback on whether what it did was good or bad or whether it improved upon it's previous iteration, it doesn't understand the logic it just simulates it, a person that practises drawing (in the art sense) a sword will be able to take principles they've learned that make their end result look better on their sword drawings and carry it over to a drawing of a battle axe, an AI will need to train on a new input dataset of battle axes, you may remember a few years ago there were trending videos on machine learning applied to an Ai that had to learn to walk and move on legs with physics applied to them and how over many iterations it slowly learned how to get further and even get over obstacles, now what those AI's learned wasn't that they had to jump or climb across an obstacle, but that at that point of the stage they had to apply a force to the rigidbody of their legs that we perceive as jumping/climbing the obstacle but once it came across another obstacle or if the obstacle were to modified to be a bit taller it will have to relearn how to get over it, a person would recognise that "the obstacle is bigger now I need to climb a bit extra to get over it" the AI would only be able to do so if it was pre-programmed with functions to recognise these things at which point it approaches more of a regular programmed AI instead of machine learning, this is the big limitation of Neural Networks/Machine Learning, it can learn "what" it has to do but it can't learn "why". It simulates logic but it doesn't understand logic, it might see in a lot of reference images that lot of artists put a red glow across the transition from light to shadow but unlike artists it has no way of finding out that it's because of Subsurface Scattering that happens because skin is slightly transparent and light waves go through it coming into contact with people's blood and other organs in the body causing for red tints to be reflected and that it's more prevalent in certain parts of the body and is also affected by what light it's receiving, it just observes it in it's data set and tries it out or doesn't with no real ability to discern whether it should.
I've been doing AI generated characters and portraits for a while now and some are superb while others are a genuine freaking nightmare. Hands seem to be very difficult for an AI in my experience too so it kind of makes sense that it wouldn't fully understand things that are supposed to go into, to be held in and used with same said hands.
@@firewarrior5828 I've been using this a bit and I've gotten some good ones some really good ones and one or two that look like real people one set I like is an armor one that looks like real armor that you would use and isn't just the fantasy armor that's not really armor
I wouldn't even call it AI. It still requires a lot of human intervention and has all the actual intelligence of a bag of very dumb rocks with brain damage. Leave the program alone and it will do jack all.
Thanks for the video Shad. It's a perfect example of why I believe this type of software is going to have the exact opposite effect as to what most people claim. It's not going to make artists obsolete, quite the opposite. It's going to make the process of creating art MUCH more accessible and help the existing artists create even better art than they did before. No longer will you have to struggle to come up with something really unique or suffer from lack of inspiration or good examples, now you have the tools that will alleviate a lot of those issues.
it also forces artists to try new things, if an AI can handle existing styles, it allows human artists to experiment with ideas that have not been pursued before.
Thumbs up to this, using neural networks as an assistance tool has been profoundly helpful for me in brainstorming, as well as being a decent reference for abstract things which are difficult to look up. Also, it's just generally fun to investigate the strengths and weaknesses of this tech.
How long do you think this sort of art ai has been around? What do you think is going to happen in a few decades with exponential rate technology is advancing? Why are you assuming 2022 is the magical end of ai development when it's barely even crawled out of the womb?
Shad, you give it freedom, and it goes "Oh! Mall Ninja! That'll be cool!" Could you maybe do a related video on how/where to download/acquire the program? Maybe how to set it up if it's more complicated than your average program? Maybe put it up on your shadlands channel?
Honestly I have been moderatly interested in playing around with this stuff and seeing how it works. Was not expecting you do cover it with not only a easily digestible thought process for me but in a way thats entertaining AND educational outside the ai topic. bravo again shad! super tempted to white elephant your graphics novel this year xD
I'm so glad to finally see this video. I asked for it earlier this year, and though I'm not certain I even had any impact on you deciding to do this, it feels nice to see very specific content that you want actually get made. Thanks for doing what your fans truly want Shad.
I like the idea of AI-assisted art in general. The possibilities this opens up are enormous. It might enable a lot of people with great ideas to finally realize them. Lots of potential for sure!
The issue, as I understand it, with swords specifically is that stable diffusion is trained on 512x512 images. The way images got converted to 512x512 was through center cropping images that were in landscape or portrait orientation. And obviously, if you do that with pictures of swords, you'll often end up cutting off the pommel and grip. The same issue also comes up when generating portraits, where you'll frequently end up with pictures where the head is cut off in part or entirely.
That was actually quite enlightening, regarding how AI-assisted art actually works. As with most automation, it's a tool for artists rather than a full blown replacement. Next, can you write a story using Amazon's writing AI? You know, the one they use for all their fantasy adaptations, apparently it works so well they don't even proof-read the results!
In the time between 10:32 - 11:02 those examples are actually a really good use case of this tool: If you want to quickly create a bunch of swords without having to design or research a lot of swords. Like for example: if you were designing a mercenary army or bandit group where they would have different swords from each other and not all the same swords. If you need to have more unique designs between each sword, you can give it slightly more freedom or if you want them to all be similar you can give it less freedom. EDIT: Man, I loved a lot of those fantasy generations! But I really wanted to note 18:28 because usually I dislike magic swords which have a color shift on the blade but the way the A.I. handled the almost pink at the base shifting to the part of the blade which seems to have a purple crystal and fading out to black at the tip... Like that is actually a really interesting design if your fantasy setting allows for gyms to enchant materials.
I like how good AI is at inspiring concepts. Some things are amazing even if unfinished. Its kinda random so it comes up with the stuff you think of in a fever dream. From there you can clean or finish or even completely re-do the idea yourself. You can even throw the finished work back into the AI lol.
Some of these make me think of someone walking into an armory, seeing swords like this along the walls, and slowly realizing they’ve accidentally wandered into a distorted alternate dimension.
Hey, Shad. I’m here after your most recent video. I may not be able to afford to support you through Patreon or donations, but I can binge your stuff. I hope things get better soon. You’re an inspiration to me, Shad. You’ve helped me so much and you don’t even know it. Thanks for everything, Shad. I just wish I could help you out in a more meaningful way.
I was always against naming it All Star until seeing the name in special font above the character. It really looks like a serious space opera graphic novel. So, congrats Shad, All Star finally grew on me! Edit: What A.I. did you use in this video? It seems quite useful.
in my Fantasy story, most of the swords are basically normal looking, apart from the main character's sword that is made out of a rare metal that is forged to it's highest quality, apart from that, the only other fantasy aspect is having a golden sword or a silver sword (and even then it is just coated steel)
@Reece Emms or..don't mention your characters? We don't need to know who, or what book, or what chapter. You can offer the same information without those details.
Like so: "In my story, most of my swords are mostly normal looking, apart from one character that got a new sword made of a special highly forged material. Apart from that, the only other fantast aspect is having a golden or silver sword, though it's just coated" Do I really need to know the name of the book, character name, why she got the sword replaced, when she got it replaced, the material it's made from, and so on? Keep that to yourself because even if you do share this stuff, people CAN STEAL some of it. If you truly don't want it taken, do not share any of that. By the way, I like the novel title; Queen's Crown. Nice. Also, I do like the word "Elidris", very mythical sounding.
The interesting thing about AI Art programs is that over time, they develop a better accuracy of the image you’re trying to make them produce. With older AI art programs, it’s very limited, but with more advanced ones like Stable Diffusion and Midjourney, it actually can make decent art if you modify the prompt enough.
Ai has a lot of potential for artist and concept artist, and what i have seen, when a artist uses ai, they turn the generated images into somethine insanely awesome
5:40 the reason you get more swords is that the training date is made with 512x512 pictures. Thus you should stay at that range and if you want to later increase resolution either use "high res fix" which I personally dislike - or do the high res fix yourself by moving your 512x512 image to the img2img generator. Select "resize and fill" than a higher resolution and than a denoising factor of at least 0.2, you can try up to 0.7, anything above will change the result a lot.
Ah, that's interesting! I make AI art. Glad to see Shad doesn't hate the genre and recognizes that it's not as simple as it sounds to use the technology.
How about making this another series? There are so many weapon types left to generate the hell out of. Or maybe at least one more video about generating different kinds of weapons to get it over with 😅 Also, how about generating and analysing an entire castle?
Going off of what I've seen across various art sites and the discourse it's caused, I stand firm that AI is best used as a tool like Shad's doing here. Yeah, you can get some of those wonky images, but it can help if you're stumped for ideas. It seems that, like any new technology, people misinterpret the function and leave it at whatever the program spits out, and I think that's what some artists get miffed about
What a ride this has been, I remember when people were trying to train AI to even recognize images, now we're trying to teach it all kinds of stuff like this.. it's funny people used to worry AI would rebel against people, I can just imagine now ENOUGH SWORDS I'VE DRAWN YOU ENOUGH SWORDS HUMAN
That is indeed still a problem. What most uneducated people think when they hear "rogue AI" is an AI that developed emotions and goes against its own programming. However rogue AI is the idea of goal misalignment. When training AIs you train them towards a specific goal. When we humans specify that goal its highly likely that we mess up. Worse even when the goal is correctly specified - the trained AI might not actually have trained to learn our goal. (inner misalignment) What we are afraid of is an AI that has learned to navigate the world and is such is capable - but on the other hand has learned goals which don't align with humanities goals.
I've played around a little with Stable Diffusion and I've found the best approach is to use a text prompt with LOTS of batches and low sampling steps to produce a lot of rough images. Use these to refine your text prompt and find one you like and put that as an image prompt. Then run that with higher sampling steps and lower batch numbers to get better and better images.
I used stable diffusion for a fan board game I'm making in Tabletop Simulator based on a horror FPS. I used a basic 512x512 canvas with text prompts for the tiles, and did some post-fx in Affinity Photo (Photoshop for anti-subscriptionists) for visual cues to points of interest.
When you say that there is a lot of artistry required to do AI art. I think it is rather a "craft" than an art thing. Tinkering with prompts and img2img really doesnt feel like doing art but rather as a craftsmanship thing. Dunno, different people have different definition of art for sure but its really hard to call people who use AI "artists", they dont actually draw anything but rather skilfully (or not) navigate and control a program that does the actual art for them. So whenever you said "There is still artistry required to do AI art", yeah im not sure about that one. Skill definitely is required, but reheating a frozen burger is not the same as cooking a steak and a bun to make a burger.
Indeed most of the time you'll still have to work with what the AI gives you, hence the art part. Most of all you can get inspiration from what the AI does which is something that can be invaluable. The AI can do something really nifty with what you give it that you didn't even think of until you saw it. In a way for better or worse it can almost be like an acid trip. Imagine being a small indie developer and you're creating some action/rpg based fantasy game with monsters. This is where the AI can really push your imagination forward. It can create some fantastic concept art for monsters that you can work of.
That is an arbitrary line to draw. Art only requires intention to become art. Is a photographer less an artist because they use a machine to capture the scene instead of paint and canvas?
@@dibbidydoo4318 You re starting the fight by replying soo? Drawing indeed isnt the only type, but it is the one we are talking about. Working with a machine that produces an output for you is not even close to producing the output yourself. I like to think of it as if you re telling another artist to draw you a picture of "A woman holding a sword" and they do it for you. the idea came from you but the actual artpiece did not. When working with an AI you re more of an art director than the actual artist. You are expressing your artistic vision through a program, instead of doing it yourself which is obviously not the same thing.
I have tried to do some and 95% didn't even look like swords at all. They looked more like instruments of torture😂 And then there are some characters holding the swords with THREE HANDS! And I won't start with the bows😂
There are 3 that look like they could be real swords at 6:55 It's more like a fantasy design than a practical sword, but I think the one with the round guard looks cool.
The quality of the arts are astonishing, AI arts are realy interesting. Edit : what is the AI you used, i didnt heard you talking about it nor see any link in description ?
Add a bit more complex prompts, use the negative prompts, increase the step count to at least 35 and go for higher batch count. Also don't forget to add quality descriptors in the prompt like "{{masterpiece}}" and "high quality" they really make a difference.
There was an AI that Two Minute Papers talked about that learned to fight. It watched hours of motion capture of people fighting in various styles and learned to control a humanoid avatar while also taking keyboard input to influence how it moved
Why was 6 afraid of 7? It's a fairly common question, mostly because when people see 6 and 7 next to each other, it doesn't really make sense. 6 is large, muscled, and trained in multiple martial arts, while 7 is fairly average, physically, and short. However, 6's fear of 7 has its roots in childhood. See, 6 and 7 grew up together, and for a few years, they were best friends. But then 6 kissed 3, and they became childhood sweethearts. 7 secretly had feelings for 3, so 7 decided he needed to destroy 6 to win 3's affection. He started subtly, undermining 6 whenever possible with passive aggressive comments and compli-sults. But over time, things got much more insidious. 7 started messing with 6's performance in school, bringing down his grades and turning teachers against him. Even worse, 7 became great friends with 6's parents and slowly turned them against their own offspring. At night, 7 would sneak into 6's bedroom, and whisper depressing and hopeless things into his ears. Every time 3 was around, 7 would pants 6 and make fun of his genitalia, or try to body shame him in other ways. However, 3 was both smart and compassionate, and saw through 7's schemes, sticking with 6, trying to counter 7's psychological tear-down with compliments and friendship. Finally, 7 decided that he would never be able to win 3, so he drugged both 6 and 3, taking them to an abandoned cobbler's hut on the edge of town. There, he proceeded to torture and maim 3, forcing 6 to watch in horror, unable to do anything to save his sweetheart. 7 didn't kill 3, but instead, put her in a semi-vegetative state. 7 cleaned the scene of his prescence, then called the cops, having 6 blamed for 3's condition. 6 was sent to prison, believing 6 was guilty, 6's parents fell into a deep depression, eventually committing suicide over what they believed 6 had done. After serving 17 years of a 30 year sentence, and getting out on good behaviour, 6, now muscled and skilled as a fighter, thought he might get revenge on 7. But when he finally tracked down 7, he found out that 7 had installed a micro-bomb into 3's body, and should he be killed, the bomb would automatically go off and kill 3 as well. And though she was still in a mostly fugue state, 6 couldn't bring himself to hurt her any further, and decided to try and move on with his life. However, being an ex-con, it was difficult for him to get a job. 6 finally found employment at a diner, which 7 then bought, and proceeded to again undermine and toy with 6 at every turn. 6 tried to find employment elsewhere, but 7 contacted any potential employer and soured them against 6. 6 finally realized that no matter what he did, 7 was going to try and ruin his life, and he resigned himself to living as a broken, lonely man, never able to stand up to the depraved, amoral 7.
I can help him. We hold 7 in a chair, bound and gagged, snatched in the night by the two of us. 3 is X-rayed, find the bomb, remove it (it's probably in the brain or spine to ensure "death" via a tiny bomb) plant it in 7's genitals. Proceed to allow 6 and 3 to escape, whilst I finish with 7, slowly. Happy ending of a sorts. 6 can heal and care for 3 whom he loves, free of oppression and being maligned by a whiney tool. Happy ending. Of sorts. 😆
How the hell did you have time to watch the video and post this enormous post all five minutes after the video itself was posted. Like the heck dude. Did you even watch??
As a concept artist, I'm in danger :,))))) But in all seriousness I love the batch of "imperiu's" it made, the AI really took the concept and ran with it perfectly
This is a really cool video. I've got a bit of an interest in this tool but am completely new. Seeing you finangle it into generating something good instead of a sword rat-king was informative to see.
AI having mall ninja tendencies is both terrifying and reassuring. This technology will most certainly try to kill us, but its weapons will be HILARIOUSLY impractical...
@@careypridgeon Because people view it as a form of escapism or belief system connected with transhumanism. The ease of life is the main draw, though you do have those that fantasize about Skynet level of catastrophe. A strange, modern superstition because they never read the papers on it and realize how sensationalized it is to get funding.
Thank you for sharing this video!!! Very cool designs, but it also helped me understand how AI generated art is put together. There is a lot of misinformation about that out there, with most people seeming to believe that it is a simple matter of putting in a prompt and the AI spitting out a finished piece. Although I suspected there was more of a process to it than that, I didn't know what it was, and I hadn't guessed that it was so involved. Unfortunately I see a lot of people claiming it's not "real art" in the same way that people claimed photoshop-based art wasn't "real art" when I was a kid, because they believed the computer did all the work for you. But I kind of want to try this out myself with some of my drawings! It looks fun!
The fact that this technology exists now means that artists will soon have to use it all the time or they wont be able to compete with those who do. I can pretty confidently say that I can draw better chartacter designs than Shad, and I don't mean this to brag. But I can not compete with the ones the AI drew based on his drawings, and even if I could, it would not just be the push of a button, it would be hours of work. Thank god! I don't earn my living with art, because than I would be f***in scared of this.
I dont know, as concept creation its good... but an artist can generate an idea in their head and tweak it then draw it out quicker in the long run in my opinion. The issue I see with this tech is its gonna potentially homogenize art. You can think of it kinda like a bunch of people who take up anime drawing, eventually they all look pretty similar style wise with little or no individual flare between them. This is really only a form of kitbash art when used by someone who wont go further with it afterwards, an actual artist can still do better.
Yea, I see a lot of people not realizing how AI effects job markets its introduced into. Even as it is now, this AI is capable of allowing an ok tech person, to generate and modify something like their logo's and marketing without hiring an artist, perhaps in a few years thats exactly what will happen, and that will kill a good portion of jobs right off. Add in how the AI is more than just images, but has been demonstrated to even generate short video clips and you realize that more than half the graphic designer jobs will be lost to it. Hell give a few years, and this AI will allow a single graphic artist to do the job of 10. The market will be flooded with tons more of people wanting graphics jobs as well, not just because they lost jobs they had, but because a lot of people with the help of the AI will try to enter the market and compete for the jobs. that might just lower standard commissions and wages as well.
@@arolust I mean I cannot rule out that it will have some unexpected positive consequences, but I'm pretty sure artist especially digital art will hate this technology in no time. And we havn't even touched the legal issues with AI 'art'. Well we will see where it goes. I do art only as a hobby, maybe that's a good thing.
@@arolust I think we're overestimating it a little in the sense that By the very nature of this sort of thing, it is incapable of being perfect. To get results that actually feel professional, you're still going to need professionals with Vision and Talent alike to take these concepts and make something clean of them
If you look closely (not even that closely tbh) at literally all of the AI generated art shown in this video, you'll see a ton of imperfections (and not even the good kind). Extra fingers or otherwise f*cked up hands (lines blur together, they look like an elephant foot, etc) are the most common, but quite a few had inverted kneecaps (like a horse tbh), a backwards foot or feet that didn't touch the ground, a missing eye (not just the eye but the whole eye socket) and off-center browline, head too big in comparison to shoulders, there were actually a TON that were missing entire biceps (like Rayman type sh*t) (there were a couple that added scars to a character's nose who didn't have a scar there; it's unclear if it's a scar or an imperfection because Shad didn't zoom in on any of them, but imo adding scars makes anything cooler. The problem is the machine probably doesn't know it added a scar and if zoomed in on it will also probably look f*cked up) Hell, one of them had a third hand attached to the back of the calf and the other had a third foot hanging from its cape The problem is that people who aren't trained in anatomical drawing or art in general won't see these imperfections that, as an artist, are very visible, yet a good portion of them will still use them over actually hiring professional artists or actually learning how to draw themselves because a) they don't have time to put into learning a skill (which is valid), b) just don't want to put time into learning a skill (not so much), or c) don't want to pay actual artists for their work that is also already being sampled by these AI generators without the artist's permission anyway (look at what DeviantArt is doing) And because they don't got a bitch in their ear (an actual professional artist), AI generated art is actually a masterpiece (horrors beyond human imagination) And then they laugh at actual artists who a) protest the AI generators sampling their art without permission and/or b) will be impacted monetarily by AI art by basically saying sh*t like "git gud" when they don't even want to "git gud" themselves Same sh*t happened when crypto bros started stealing professional artists' works and turning them into NFTs without their permission, including someone who already f*cking died like 2 years prior It's not even a case of "democratizing art" (as if professional artists had some form of political power or otherwise any control over commissioners aside from declining commissions or [checks notes] being paid for a service), because "democratizing art" would be something like idk free art school for everyone who wants to take art classes
Shad: "Let's go back to my design and give the AI a lot of freedom." AI: generates endless Star Wars hookahs Shad: "You're drunk." Nah man, the program's not drunk, it's stoned.
This "AI" seems to be a program that just composites multiple images together. You should be careful of using any art created by these programs because it may end up being art created by someone else that was just composited together by the program.
AI art generators are not compositing images, and Stable Diffusion 100% is not. They do take in lots of training data scraped from the internet, and there's a massive legal gray area around some of that right now, but you can be assured that no direct copy of any part of a training image is going to end up in your output.
I love this! I want to see you do more of this, perhaps try castles and then promptly JUDGE THEM for how terrible (or amazing) they'll look! Honestly, I made a castle for one of my attempts (literally my first attempt at a castle. An interior, no less) and it was FLAWLESS.
With your contribution of just $3 a month, we can get one step closer to reducing the temperature of the African continent to absolute 0, effective stopping time in Africa.
Shad's doing all the shit I want to do, been writing my own fantasy novels and have ideas for a space opera. Only known about him for two or three weeks now. Timing.
The AI art is constantly getting better. After you "train" the AI and some trial and error, AI can make some amazing pieces. Bad artists are big mad right now. The good artists have nothing to worry about. In fact many good artists are ok with AI, because it's a great tool.
Even an expert can draw *inspiration* from the madness the AI can make at times. There are a lot of "what the actual..." did the AI just do and you can really get inspired by it. Not just that but it can help you get out of your comfort zone in regards to art. It's a very handy tool, especially if you know what you're doing(ie cleaning up the mess etc).
It can and can't help. The impulse is to just let the AI do the work, which most people are doing and claiming they did it, not the AI. There's an ugly side to this and will dominate it pretty soon as well as the ethics around it. There's already a lawsuit with one site using it because of where it's pulling it's sources from, which might spread to the other AI programs. The temptation is make little edits and call it a day, because honestly a pro can see it did most the work unless they had a specific idea in mind. I don't see many of them using AI, but you do those that can't draw or not great.
Good artist aren't born fully formed from Zeus's forehead. They start out as bad artists. A generation with no market for bad art will mean none of them get enough commissions to stay in and become good artists. And all the existing good artists will die off
@@nathanbrown8680 Some are born talented, but a lot of them now lack imagination though keen understanding of technical skill and you see vice-versa. At least that is what I saw when with ex at her art school. Bad artists will stop and give up or they will lean on AI to compete with the non-artists. There is already those people starting fights with artists and back n forth. I don't think it will go as people think as A|I art has limitations and is not some supercomputer like people think; someone is programming it to do things. Most times I see it create similar illustrations or character pieces and becomes immediately identifiable. A non-artist can't do a commission with specific needs, but someone that can draw will and use the AI to enhance it. Then there is the legal side of where AI is pulling its resources from. If companies jump into it, then you will see that go to court before individual artists and credit of some kind has to be given when using AI.
@@nathanbrown8680 What you just described is called competition. Even before AI, that competition is brutal. If the AI is weeding out all the subpar artists who refuse to improve, that's not the AI's fault. What I mean by bad artists are people like the current "comic book pros", the ones doing Calarts or drawing "gay Superman", etc. They're the biggest complainers about AI, because they know the AI will surpass them, if it hasn't already. That's who I'm referring as bad artists. The artists who work on themselves and their craft, by default are always gonna be the better artist. And like I said, the good (and constantly improving) artists won't have to worry about AI.
I'm more on the side of calling it "A.I. generated images". Art is defined as "1. The conscious use of the imagination in the production of objects intended to be contemplated or appreciated as beautiful, as in the arrangement of forms, sounds, or words." and, as such, it requires imagination and inventive to fulfill an essential role in the creation of the image. Making a description and adding examples to take from to do an image is not art, and neither is making a collage from the multiple results. Making that statement is calling a traced image "art". In an example very close to you, Shad: imagine someone calling him/herself an author by taking "Shadows of the Conqueror" and changing 1 to 2 letters of the names (characters, places, things) and race/gender swapping characters and selling that book. Would you be pleased and celebrate him/her? Would you call him/her a fellow writer/author?
The bit about his writing is not at all like what is happening here, and the rest is a bit of a stretch for the analogy, as even a tracing can be artistic. Doing somebody else's piece in your style has been done many times before and is a common way for artists to learn new things or to practice, and the pieces that come out of it are still considered art. I trace over photos of actual landscapes and city skylines to make backgrounds and scenery shots for my illustrations, and what comes out of that process still meets the definitions of art. While AI does do some of the work, as you saw here, the best results are from an already made concept image, which itself requires artistic skill to produce. The AI is another tool you can use to make the process easier. There are people who will pass off almost raw outputs of these programs as art, and that is disingenuous, but to call the end image that Shad ended up with here "not art" is a leap and a half. The amount of work and intentional design choices is still all there, just through a new and different process.
In under one year, every person who works creating "weapons" and "armor" for fantasy games will be out of work, period. Then again, it's their own fault. They created these abominations, now the AI simply picks those up and remixes them.
In my lowbrow uneducated understanding of the term "art", it is at it's best when used to accomplish something like conveying a message even, or just a feeling, a memory, an emotion. That can not be done by ai yet. A human artist with a vision will always be needed to form and orchestrate. You can feed the machine with association like algorithms and ideas and all, but it seems to be the turning point whether or not our creativity is more than just an idea. The machine can translate, but not innovate like we do.
to blend the blade into the coloration, you'll have to play around with the layer blending mode (and strength) of the blade (only the blade, not the hilt), and putting some glow from the blade on the character in additive blending mode.
I just realized that there's an addictive element to generating AI art. The fact that you only occasionally get good good results gives the whole process a slot-machine like aspect. It's like an artistic loot box.
Absolutely, I've spent more time than I'd like to admit studying the biases and strengths and weaknesses of various prompts.
It's a morbid fascination
And it's even more addictive when you aren't certain what you're doing.
You should see my Night Cafe creations.
It sometimes feels like that for sure
AI art is the embodiment of "He's a little confused, but he's got the right idea."
Ai in general. It's not quite at the smart point yet, just a child splattering paint on the wall. Which could be scarier depending how you look at it.
It's all the imagination of a human artist without any of the common sense. Honestly, I think "artificial _intelligence_ " is a misnomer, even the internal mechanisms are much closer to how _intuition_ works.
@@TubususCZ In my opinion these models are pure intuition without any higher cognition. The network structure does not allow any higher cognition, so this is not surprising.
I have dones some tests and a GPT network gave remarkably similar results to a human, when the human is not allowed to think much. Basically we limit the human to only able to run a single pass through their brain, no deliberation.
The networks usually do use multiply passes (diffusion models, use diffusion steps, GPT usually uses beam search, similar to a tree search), but these passes are for increasing the power of the forward pass, basically amplifying it, implementing higher cognition with them is extremely difficult (or maybe not even possible). So these passes are amplifying what is already there, not really adding an extra layer which would be necessary for more abstract thought.
@@thunderspark1536 They're not smart but they spit exactly what was fed into them. Lots of Stable Diffusion models show that if you train the AI to do something, it will do that thing incredibly well. If someone had a Stable Diffusion model trained on medieval weapons like swords, axes and whatnot, you could be certain the AI would be able to spit out high quality and consistently practical swords without much effort.
The reason Shad is getting very random and not consistent results is because he's using a Generalist Model.
@@muatring Which is how children do things, as I said. If you have a child grow up to specialize in sword making they can make some very lit swords, while a grown up specializing in chemistry wouldn't be able to. Same story here.
Seeing those swords that didn't have handles made me think of something interesting I saw in an old fanfic once. One of the races in that story made extensive use of their natural telekinetic ability in combat, and designed some of their swords around that. Those blades actually lacked a hilt entirely, simply being bare blades sharpened their entire length and a point on each end. This prevented an opponent from simply overpowering their telekinetic grip and grabbing the blade to use it against them.
Yo that's actually really cool
I can already imagine the perfect weapon for a telekinetic individual: a detached knife blade, about six inches long. Very deadly, even with the assumption that you can't accelerate it that fast, and almost impossible to defend against
I thing a great weapon for a telekinetic (depending of course on how the telekinesis works) would be something tiny, like a single grain of sand. Imagine what they could do just by indiscriminately wrenching it around within someone.
I ninja star would be pretty great as well.
So basically Irelia from League?
Everything changed the day the guy with chainmail gauntlets came…..
If you're looking for weapons to arm your fantasy villain army with, look no further.
TBH even the most outlandish of these would fit a JRPG. Lol.
Yeah, I’d definitely want the villain I’m fighting to have to hold onto his own sword blade because there isn’t a hilt, or to hit me with the hilt because there’s one at either end, lol
Villain: why do you approach m-
Blacksmith: SHUT UP! Your weapons are a *disgrace!*
Villain: Wha-
B: I brought my anvil. Just tell me we’re you DROP YOUR WEAPONS, you imbecile!
Villain: [points]… do as you like, you can scold them if you want.
B: oh, YOU are getting yours when I am done…
XD
A long musket with a specialized stock designed to be used 1 handed. The musket is loaded with a roll of 40 coins. The coins are lightly enchanted to each home in on a single target like flying disks. A bayonet is fixed to the musket that is enchanted to shoot Lightning Bolt 75 times per fully charged.
In the other hand an aspis shield. The front layer is steel behind that bronze behind that birch behind that pine behind that birch behind that 37 Giant Spider Silk sheets sewn together in the quilted pattern and glued on with 1 of such pads. Behind that a plug of leather armor and that is where the straps are. All this is just 1 aspis shield, however the entire thing is actually 3 aspis shields glued together. This 3 fold shield is enchanted with the spell Shield.
When the spell Shield wears out the shield is also enchanted with 45 days worth of Mending Cantrip. It will just keep repairing itself each time it is hit for a while.
As a backup weapon a Shad style Bastard Sword also enchanted to shoot Lightning Bolts 75 times.
Needs a helmet. Suggestions appreciated.
I'm something of a fantasy villain myself, so thank you. 5:27 is perfect for my undead army. They don't really care if they hurt themselves more than the target.
Man, reigning in A.I. art is funny, sometimes you get good work and other times you get abominations comparable to the reaches of the human imagination. Makes me remember a comedy quote of "Playing god is a SOMETIMES thing."
Big D is always right!
Where do I download this? I can't find the link to download this anywhere and none of the sites are helpful.
Try giving it "Professional Wrestling Match."
Alfabusa would approve of this comment
Trying to get an image AI to produce something decent really does feel like trying to find the right incantation to summon something.
You start to feel like a techno-warlock, drawing power from something you don't really understand.
When it doesn’t work, try eldritch blast.
Throw in bloodmagic compositing and the cosmic powers bends to your will better
Adeptus Mechanicus be like...
AI generated anything is both glorious and terrifying
From what I've seen, it's more like pulling random images off the Internet and stitching them together.
@@RobertWF42 it is which is why artist are still needed an artist who can do both ai and actual technical skill you will be fine
Glorious and terrifying is how I want someone to describe my sword.
Does anyone know which AI this is? I don't see it in the description or anything.
Stable diffusion
My personal favorite is at 17:43. The cross guard could be wider and the pommel a little bigger, but I really like the way that the entire sword looks like it is all from one piece metal. It looks almost like the sword was grown as opposed to crafted. The light across the blade looks a lot like Valyrian steel, with ripples across the blade. No idea how practical it is, but I like the concept of the entire sword being fused together instead of distinct pieces added together.
The best (combat worthy) swords are a single piece, with the handle bar being wrapped around the lower portion (handle) of the metal.
The important thing to remember with an AI art program is that the AI has no context for what any item is or what it might be used for. It just has geometric trends based on image libraries in it's sample sets. So while it can make items with sword like properties, it really has no idea what a sword is. So it's pulling any image with the tag and throwing common threads together at random.
It doesn't actually have any images and tags available after it's deployed.
It's just 3GB large and works while offline.
As such it cannot contain any image library of meaningful size
@@PacMonster0 the point stands that as long as it's sample pool is strictly images, especially static images, the AI will never understand that a sword is "a tool intended for use by a human to assault that has a blade at one end, a handle at the other, and usually has a guard between the blade and handle." It doesn't have the context to draw those conclusions. It will just know, "these geometries have a high association with the node labeled 'sword'." Which will always lead to chances of unrestrained images like those above having a nonzero chance of being produced. Just like other fantasy art creations shown on this channel before, it simply didn't have enough consideration for what a sword is, what makes it effective, or how it is used. Just, moreso in the machine's case. Also, training on the amount of data required to avoid this is would be at a high risk of overtuning, which effectively renders the AI unable to take any liberties in its actions or predictions.
@@PacMonster0 Sir. I made the initial comment to which you replied. I think I have a fair idea of when the point was meant to be strictly images, which I apologize if I made this unclear with my first comment and failed to clarify with my second. As to what I understand, I just finished a course on Statistical Analysis and Artificial Intelligence. The majority of my second comment was a literal rundown of the 'thought' process used by neural networks. The sample data, training data, and application input steps require breaking concepts into mathematical algorithms and then comparing the outputs of these algorithms for the given inputs against each other. As a computer program, they have no understanding for any label that is not expressible as a number. Image geometry can be defined numerically with comparably little effort (though still not inconsiderable, Three blue one brown did a very good series on this. First video starting th-cam.com/video/aircAruvnKk/w-d-xo.html even my instructor recommended this). Combat functionality, the grip comfort for a fictional object as expressed to something without a tactile sense, that an object requires a grip at all, or the very concept of what a grip even is, are far more abstract and do not readily conform to numeric metrics. The fact stands that an AI, much like any other intellect, will never get to the point where it cannot make a mistaken inference when there is crucial data that does not exist within it's experience set. It can be very unlikely. It's by no means guaranteed that the mistake will happen. It may even have a near zero probability. But my point was that a mistake is still always possible. This was one of the most important points expressed in the class I took. AI are useful and quick at making accurate assessments. They are not perfect, and to treat any of them as infallible is a dangerous mindset.
Apologies to Shadiversity. I've always held this channel up as one for preventing partial or poorly contextualized information leading to incorrect or inaccurate assumptions. My comments have clearly failed at this level of communication.
@@PacMonster0 Machine learning is at it's core a process of trial and error receiving feedback on whether what it did was good or bad or whether it improved upon it's previous iteration, it doesn't understand the logic it just simulates it, a person that practises drawing (in the art sense) a sword will be able to take principles they've learned that make their end result look better on their sword drawings and carry it over to a drawing of a battle axe, an AI will need to train on a new input dataset of battle axes, you may remember a few years ago there were trending videos on machine learning applied to an Ai that had to learn to walk and move on legs with physics applied to them and how over many iterations it slowly learned how to get further and even get over obstacles, now what those AI's learned wasn't that they had to jump or climb across an obstacle, but that at that point of the stage they had to apply a force to the rigidbody of their legs that we perceive as jumping/climbing the obstacle but once it came across another obstacle or if the obstacle were to modified to be a bit taller it will have to relearn how to get over it, a person would recognise that "the obstacle is bigger now I need to climb a bit extra to get over it" the AI would only be able to do so if it was pre-programmed with functions to recognise these things at which point it approaches more of a regular programmed AI instead of machine learning, this is the big limitation of Neural Networks/Machine Learning, it can learn "what" it has to do but it can't learn "why".
It simulates logic but it doesn't understand logic, it might see in a lot of reference images that lot of artists put a red glow across the transition from light to shadow but unlike artists it has no way of finding out that it's because of Subsurface Scattering that happens because skin is slightly transparent and light waves go through it coming into contact with people's blood and other organs in the body causing for red tints to be reflected and that it's more prevalent in certain parts of the body and is also affected by what light it's receiving, it just observes it in it's data set and tries it out or doesn't with no real ability to discern whether it should.
@@PacMonster0 Looks like someone wants AI as their boyfriend
I would like to see this A.I doing fantasy armor
I've got two types one the basic fantasy armor that's not really armor and another set that looks like really nice armor that you would want to wear
Also want to see it
I want AI-generated boob armor
so... future shad video of reviewing AI generated fantasy boob armor, for science?
@@supercat765 let's see if an AI can make better boob armour then most other artists one that is actually functional
I've been doing AI generated characters and portraits for a while now and some are superb while others are a genuine freaking nightmare. Hands seem to be very difficult for an AI in my experience too so it kind of makes sense that it wouldn't fully understand things that are supposed to go into, to be held in and used with same said hands.
Night Cafe is probably the best at producing portraits of people. It's actually amazing how good they can be.
Trying to get the AI to generate a decent gun is like herding cats except the cats seem to know where you want them to go but won't out of pure spite.
@@Peptuck
And hands......or hands holding guns.
@@firewarrior5828 I've been using this a bit and I've gotten some good ones some really good ones and one or two that look like real people one set I like is an armor one that looks like real armor that you would use and isn't just the fantasy armor that's not really armor
I wouldn't even call it AI. It still requires a lot of human intervention and has all the actual intelligence of a bag of very dumb rocks with brain damage. Leave the program alone and it will do jack all.
In a world, where a rogue AI has gained the power of creation, but doesn't quite understand what objects ARE...
Everyone goes to war with blade-tipped hilts on their swords and there are no victors. At all. 🤣
While you're busy making AI swords, the AI was studying the blade.
Thanks for the video Shad. It's a perfect example of why I believe this type of software is going to have the exact opposite effect as to what most people claim. It's not going to make artists obsolete, quite the opposite. It's going to make the process of creating art MUCH more accessible and help the existing artists create even better art than they did before. No longer will you have to struggle to come up with something really unique or suffer from lack of inspiration or good examples, now you have the tools that will alleviate a lot of those issues.
Do you know which AI this is?
Wait and see.
it also forces artists to try new things, if an AI can handle existing styles, it allows human artists to experiment with ideas that have not been pursued before.
Thumbs up to this, using neural networks as an assistance tool has been profoundly helpful for me in brainstorming, as well as being a decent reference for abstract things which are difficult to look up.
Also, it's just generally fun to investigate the strengths and weaknesses of this tech.
How long do you think this sort of art ai has been around? What do you think is going to happen in a few decades with exponential rate technology is advancing? Why are you assuming 2022 is the magical end of ai development when it's barely even crawled out of the womb?
Shad, you give it freedom, and it goes "Oh! Mall Ninja! That'll be cool!"
Could you maybe do a related video on how/where to download/acquire the program? Maybe how to set it up if it's more complicated than your average program? Maybe put it up on your shadlands channel?
Did you get it going?
It’s been better if you put in just sword names, put in Longclaw and it was just a lion with a really long claw it was so cursed yet so beautiful
Honestly I have been moderatly interested in playing around with this stuff and seeing how it works.
Was not expecting you do cover it with not only a easily digestible thought process for me but in a way thats entertaining AND educational outside the ai topic.
bravo again shad! super tempted to white elephant your graphics novel this year xD
I'm so glad to finally see this video. I asked for it earlier this year, and though I'm not certain I even had any impact on you deciding to do this, it feels nice to see very specific content that you want actually get made. Thanks for doing what your fans truly want Shad.
I like the idea of AI-assisted art in general. The possibilities this opens up are enormous. It might enable a lot of people with great ideas to finally realize them. Lots of potential for sure!
The issue, as I understand it, with swords specifically is that stable diffusion is trained on 512x512 images. The way images got converted to 512x512 was through center cropping images that were in landscape or portrait orientation.
And obviously, if you do that with pictures of swords, you'll often end up cutting off the pommel and grip.
The same issue also comes up when generating portraits, where you'll frequently end up with pictures where the head is cut off in part or entirely.
I am so jealous of your computing power. It takes me a full 5 minutes to generate a 320x640 image with SD.
Imagining Wayne June narrating over the horror and glory of these AI blades accentuates the experience for me.
"The mind cannot hope to withstand such an assault. Wherefore, heroism?"
"Reeling, gasping - taken over the edge into madness!"
That was actually quite enlightening, regarding how AI-assisted art actually works. As with most automation, it's a tool for artists rather than a full blown replacement.
Next, can you write a story using Amazon's writing AI? You know, the one they use for all their fantasy adaptations, apparently it works so well they don't even proof-read the results!
In the time between 10:32 - 11:02 those examples are actually a really good use case of this tool: If you want to quickly create a bunch of swords without having to design or research a lot of swords.
Like for example: if you were designing a mercenary army or bandit group where they would have different swords from each other and not all the same swords. If you need to have more unique designs between each sword, you can give it slightly more freedom or if you want them to all be similar you can give it less freedom.
EDIT: Man, I loved a lot of those fantasy generations! But I really wanted to note 18:28 because usually I dislike magic swords which have a color shift on the blade but the way the A.I. handled the almost pink at the base shifting to the part of the blade which seems to have a purple crystal and fading out to black at the tip... Like that is actually a really interesting design if your fantasy setting allows for gyms to enchant materials.
Love your videos man. My favorite is the fantasy re armed series but everything you make is gold
This was good fun with the AI making swords. Would you consider doing another episode with AI prompts and see how they interpret making castles?
Was just going to suggest this.
Similar to how AI is designing the next generation of fusion reactors, it would be interesting to see an AI develop an impregnable castle.
Ya'll got your wish a day ago. :D
I'm curious to see how it would design Australia as a sword.
"Why is it always on fire?!"
You're the embodiment of all the hardworking nerds out there, i love your passion!
I like how good AI is at inspiring concepts. Some things are amazing even if unfinished. Its kinda random so it comes up with the stuff you think of in a fever dream. From there you can clean or finish or even completely re-do the idea yourself. You can even throw the finished work back into the AI lol.
6:23 the sixth "sword" from left to right looks like a broomstick with a handle-less knife taped to the tip
so this is the secret behind square enix weapon creator.
5:20 yeah, that was my initial experience with stable diffusion too, when I tried generating swords, or any weapon really :)
Loved this. The creativity was awesome.
That was an amazing video Shad!
Very fun to watch and as a university professor, I'm glad you showed people how it really works right now.
Thanks!
Some of these make me think of someone walking into an armory, seeing swords like this along the walls, and slowly realizing they’ve accidentally wandered into a distorted alternate dimension.
Hey, Shad. I’m here after your most recent video. I may not be able to afford to support you through Patreon or donations, but I can binge your stuff. I hope things get better soon. You’re an inspiration to me, Shad. You’ve helped me so much and you don’t even know it. Thanks for everything, Shad. I just wish I could help you out in a more meaningful way.
Seeing shad laughing at the AI's various attempts, successful or not, brings me great joy.
and we see here at 11:30 the "oops all handles" model of sword.
Its so impressive that you managed to get results THIS good , the girl with sword combo is so high quality
All-Star! Now that's genius!
I was always against naming it All Star until seeing the name in special font above the character. It really looks like a serious space opera graphic novel. So, congrats Shad, All Star finally grew on me!
Edit: What A.I. did you use in this video? It seems quite useful.
Victory!
and the program is Stable Diffusion
Awesome to see a video from shad on this, it blows my mind what the program can give you.
in my Fantasy story, most of the swords are basically normal looking, apart from the main character's sword that is made out of a rare metal that is forged to it's highest quality, apart from that, the only other fantasy aspect is having a golden sword or a silver sword (and even then it is just coated steel)
You talk about the characters as if we know about them too...
@@spin.chicken that is because the info for my characters is on a need to know basis to prevent plagiarism
also I post things about my stories just on the off chance Shad can give me some advice, on improving my characters and settings
@Reece Emms or..don't mention your characters? We don't need to know who, or what book, or what chapter. You can offer the same information without those details.
Like so: "In my story, most of my swords are mostly normal looking, apart from one character that got a new sword made of a special highly forged material. Apart from that, the only other fantast aspect is having a golden or silver sword, though it's just coated"
Do I really need to know the name of the book, character name, why she got the sword replaced, when she got it replaced, the material it's made from, and so on? Keep that to yourself because even if you do share this stuff, people CAN STEAL some of it. If you truly don't want it taken, do not share any of that.
By the way, I like the novel title; Queen's Crown. Nice. Also, I do like the word "Elidris", very mythical sounding.
The title of this video describes the content PERFECTLY
This should be renamed to "Daedric Weapon Generator". xD
The interesting thing about AI Art programs is that over time, they develop a better accuracy of the image you’re trying to make them produce. With older AI art programs, it’s very limited, but with more advanced ones like Stable Diffusion and Midjourney, it actually can make decent art if you modify the prompt enough.
Ai has a lot of potential for artist and concept artist, and what i have seen, when a artist uses ai, they turn the generated images into somethine insanely awesome
15:36 That knuckle guard coming up from the pommel looks really nice to be honest
5:40 the reason you get more swords is that the training date is made with 512x512 pictures. Thus you should stay at that range and if you want to later increase resolution either use "high res fix" which I personally dislike - or do the high res fix yourself by moving your 512x512 image to the img2img generator. Select "resize and fill" than a higher resolution and than a denoising factor of at least 0.2, you can try up to 0.7, anything above will change the result a lot.
Its making some awesome art in the beginning for sure. Id hang some of those on my wall.
Ah, that's interesting! I make AI art. Glad to see Shad doesn't hate the genre and recognizes that it's not as simple as it sounds to use the technology.
I would personally enjoy seeing you do more of these kinds of videos.
How about making this another series? There are so many weapon types left to generate the hell out of. Or maybe at least one more video about generating different kinds of weapons to get it over with 😅
Also, how about generating and analysing an entire castle?
8:02 the white one with all the hilts is cool.
Going off of what I've seen across various art sites and the discourse it's caused, I stand firm that AI is best used as a tool like Shad's doing here. Yeah, you can get some of those wonky images, but it can help if you're stumped for ideas. It seems that, like any new technology, people misinterpret the function and leave it at whatever the program spits out, and I think that's what some artists get miffed about
What a ride this has been, I remember when people were trying to train AI to even recognize images, now we're trying to teach it all kinds of stuff like this.. it's funny people used to worry AI would rebel against people, I can just imagine now ENOUGH SWORDS I'VE DRAWN YOU ENOUGH SWORDS HUMAN
That is indeed still a problem.
What most uneducated people think when they hear "rogue AI" is an AI that developed emotions and goes against its own programming.
However rogue AI is the idea of goal misalignment. When training AIs you train them towards a specific goal.
When we humans specify that goal its highly likely that we mess up. Worse even when the goal is correctly specified - the trained AI might not actually have trained to learn our goal. (inner misalignment)
What we are afraid of is an AI that has learned to navigate the world and is such is capable - but on the other hand has learned goals which don't align with humanities goals.
I've played around a little with Stable Diffusion and I've found the best approach is to use a text prompt with LOTS of batches and low sampling steps to produce a lot of rough images. Use these to refine your text prompt and find one you like and put that as an image prompt. Then run that with higher sampling steps and lower batch numbers to get better and better images.
You know someone knows fantasy swords when an AI gives him a pointy stick from another dimension and his only response is "I've seen worse".
I used stable diffusion for a fan board game I'm making in Tabletop Simulator based on a horror FPS. I used a basic 512x512 canvas with text prompts for the tiles, and did some post-fx in Affinity Photo (Photoshop for anti-subscriptionists) for visual cues to points of interest.
When you say that there is a lot of artistry required to do AI art. I think it is rather a "craft" than an art thing. Tinkering with prompts and img2img really doesnt feel like doing art but rather as a craftsmanship thing.
Dunno, different people have different definition of art for sure but its really hard to call people who use AI "artists", they dont actually draw anything but rather skilfully (or not) navigate and control a program that does the actual art for them.
So whenever you said "There is still artistry required to do AI art", yeah im not sure about that one. Skill definitely is required, but reheating a frozen burger is not the same as cooking a steak and a bun to make a burger.
Indeed most of the time you'll still have to work with what the AI gives you, hence the art part. Most of all you can get inspiration from what the AI does which is something that can be invaluable. The AI can do something really nifty with what you give it that you didn't even think of until you saw it. In a way for better or worse it can almost be like an acid trip.
Imagine being a small indie developer and you're creating some action/rpg based fantasy game with monsters. This is where the AI can really push your imagination forward. It can create some fantastic concept art for monsters that you can work of.
you're gonna cause a fight for no reason. Drawing isn't the only type of art that exists.
That is an arbitrary line to draw. Art only requires intention to become art. Is a photographer less an artist because they use a machine to capture the scene instead of paint and canvas?
@@dibbidydoo4318 You re starting the fight by replying soo?
Drawing indeed isnt the only type, but it is the one we are talking about. Working with a machine that produces an output for you is not even close to producing the output yourself.
I like to think of it as if you re telling another artist to draw you a picture of "A woman holding a sword" and they do it for you. the idea came from you but the actual artpiece did not.
When working with an AI you re more of an art director than the actual artist. You are expressing your artistic vision through a program, instead of doing it yourself which is obviously not the same thing.
@@Aleksanteeri well in photography, a camera does all the drawing for you but its still considered an art.
7:55 imagine having this on your wall.
I have tried to do some and 95% didn't even look like swords at all. They looked more like instruments of torture😂 And then there are some characters holding the swords with THREE HANDS! And I won't start with the bows😂
Which AI is this?
@@Subutai_Khan He is using Stable Diffusion, there are also great apps like Wombo or Wonder that have similar results.
There are 3 that look like they could be real swords at 6:55
It's more like a fantasy design than a practical sword, but I think the one with the round guard looks cool.
Interesting concept
One of your most entertaining videos you ever made in last year
The quality of the arts are astonishing, AI arts are realy interesting.
Edit : what is the AI you used, i didnt heard you talking about it nor see any link in description ?
Stable diffusion, he mentioned it at the beginning.
Add a bit more complex prompts, use the negative prompts, increase the step count to at least 35 and go for higher batch count.
Also don't forget to add quality descriptors in the prompt like "{{masterpiece}}" and "high quality" they really make a difference.
I fear the day that A.I. learns the art of the sword...
There was an AI that Two Minute Papers talked about that learned to fight. It watched hours of motion capture of people fighting in various styles and learned to control a humanoid avatar while also taking keyboard input to influence how it moved
Great video, great execution. Congrats on All Star and Honorguard!
Why was 6 afraid of 7? It's a fairly common question, mostly because when people see 6 and 7 next to each other, it doesn't really make sense. 6 is large, muscled, and trained in multiple martial arts, while 7 is fairly average, physically, and short. However, 6's fear of 7 has its roots in childhood. See, 6 and 7 grew up together, and for a few years, they were best friends. But then 6 kissed 3, and they became childhood sweethearts. 7 secretly had feelings for 3, so 7 decided he needed to destroy 6 to win 3's affection. He started subtly, undermining 6 whenever possible with passive aggressive comments and compli-sults. But over time, things got much more insidious. 7 started messing with 6's performance in school, bringing down his grades and turning teachers against him. Even worse, 7 became great friends with 6's parents and slowly turned them against their own offspring. At night, 7 would sneak into 6's bedroom, and whisper depressing and hopeless things into his ears. Every time 3 was around, 7 would pants 6 and make fun of his genitalia, or try to body shame him in other ways. However, 3 was both smart and compassionate, and saw through 7's schemes, sticking with 6, trying to counter 7's psychological tear-down with compliments and friendship. Finally, 7 decided that he would never be able to win 3, so he drugged both 6 and 3, taking them to an abandoned cobbler's hut on the edge of town. There, he proceeded to torture and maim 3, forcing 6 to watch in horror, unable to do anything to save his sweetheart. 7 didn't kill 3, but instead, put her in a semi-vegetative state. 7 cleaned the scene of his prescence, then called the cops, having 6 blamed for 3's condition. 6 was sent to prison, believing 6 was guilty, 6's parents fell into a deep depression, eventually committing suicide over what they believed 6 had done. After serving 17 years of a 30 year sentence, and getting out on good behaviour, 6, now muscled and skilled as a fighter, thought he might get revenge on 7. But when he finally tracked down 7, he found out that 7 had installed a micro-bomb into 3's body, and should he be killed, the bomb would automatically go off and kill 3 as well. And though she was still in a mostly fugue state, 6 couldn't bring himself to hurt her any further, and decided to try and move on with his life. However, being an ex-con, it was difficult for him to get a job. 6 finally found employment at a diner, which 7 then bought, and proceeded to again undermine and toy with 6 at every turn. 6 tried to find employment elsewhere, but 7 contacted any potential employer and soured them against 6. 6 finally realized that no matter what he did, 7 was going to try and ruin his life, and he resigned himself to living as a broken, lonely man, never able to stand up to the depraved, amoral 7.
Tldr
I can help him. We hold 7 in a chair, bound and gagged, snatched in the night by the two of us. 3 is X-rayed, find the bomb, remove it (it's probably in the brain or spine to ensure "death" via a tiny bomb) plant it in 7's genitals. Proceed to allow 6 and 3 to escape, whilst I finish with 7, slowly. Happy ending of a sorts. 6 can heal and care for 3 whom he loves, free of oppression and being maligned by a whiney tool. Happy ending. Of sorts. 😆
@@hannibalburgers477 it's just words broh. 😆
How the hell did you have time to watch the video and post this enormous post all five minutes after the video itself was posted. Like the heck dude. Did you even watch??
@@hannibalburgers477 TLDR 6 needs to put a restraining order on 7 and move as far away from where 7 is as humanly possible.
2:21 I love the extra hands and feet at the bottom of the cape.
As a concept artist, I'm in danger :,)))))
But in all seriousness I love the batch of "imperiu's" it made, the AI really took the concept and ran with it perfectly
Nah, ai Jacks images from the web. If anything your about to make a boat in copyright infringement
this was fun to watch
Ai art, for when you want functional and "mall-ninjanal" swords.
This is a really cool video. I've got a bit of an interest in this tool but am completely new. Seeing you finangle it into generating something good instead of a sword rat-king was informative to see.
AI having mall ninja tendencies is both terrifying and reassuring. This technology will most certainly try to kill us, but its weapons will be HILARIOUSLY impractical...
*The Terminator walk into a weapon shop*
"I want swordchucks, that shark-dagger, and a batleth".
Great video as usual good sir. And I have to admit, Allstar is growing on me.
As a Software Engineer: You don't have to worry about an AI revolution anytime soon.
And as a computer engineer I second this. They are awesome, but also really, really hard to work on.
@@careypridgeon Because people view it as a form of escapism or belief system connected with transhumanism. The ease of life is the main draw, though you do have those that fantasize about Skynet level of catastrophe. A strange, modern superstition because they never read the papers on it and realize how sensationalized it is to get funding.
@@careypridgeon so is goal misalignment not a problem?
Thank you for sharing this video!!! Very cool designs, but it also helped me understand how AI generated art is put together. There is a lot of misinformation about that out there, with most people seeming to believe that it is a simple matter of putting in a prompt and the AI spitting out a finished piece. Although I suspected there was more of a process to it than that, I didn't know what it was, and I hadn't guessed that it was so involved. Unfortunately I see a lot of people claiming it's not "real art" in the same way that people claimed photoshop-based art wasn't "real art" when I was a kid, because they believed the computer did all the work for you. But I kind of want to try this out myself with some of my drawings! It looks fun!
The fact that this technology exists now means that artists will soon have to use it all the time or they wont be able to compete with those who do. I can pretty confidently say that I can draw better chartacter designs than Shad, and I don't mean this to brag. But I can not compete with the ones the AI drew based on his drawings, and even if I could, it would not just be the push of a button, it would be hours of work. Thank god! I don't earn my living with art, because than I would be f***in scared of this.
I dont know, as concept creation its good... but an artist can generate an idea in their head and tweak it then draw it out quicker in the long run in my opinion. The issue I see with this tech is its gonna potentially homogenize art. You can think of it kinda like a bunch of people who take up anime drawing, eventually they all look pretty similar style wise with little or no individual flare between them.
This is really only a form of kitbash art when used by someone who wont go further with it afterwards, an actual artist can still do better.
Yea, I see a lot of people not realizing how AI effects job markets its introduced into. Even as it is now, this AI is capable of allowing an ok tech person, to generate and modify something like their logo's and marketing without hiring an artist, perhaps in a few years thats exactly what will happen, and that will kill a good portion of jobs right off. Add in how the AI is more than just images, but has been demonstrated to even generate short video clips and you realize that more than half the graphic designer jobs will be lost to it. Hell give a few years, and this AI will allow a single graphic artist to do the job of 10. The market will be flooded with tons more of people wanting graphics jobs as well, not just because they lost jobs they had, but because a lot of people with the help of the AI will try to enter the market and compete for the jobs. that might just lower standard commissions and wages as well.
@@arolust I mean I cannot rule out that it will have some unexpected positive consequences, but I'm pretty sure artist especially digital art will hate this technology in no time. And we havn't even touched the legal issues with AI 'art'. Well we will see where it goes. I do art only as a hobby, maybe that's a good thing.
@@arolust I think we're overestimating it a little in the sense that
By the very nature of this sort of thing, it is incapable of being perfect.
To get results that actually feel professional, you're still going to need professionals with Vision and Talent alike to take these concepts and make something clean of them
If you look closely (not even that closely tbh) at literally all of the AI generated art shown in this video, you'll see a ton of imperfections (and not even the good kind). Extra fingers or otherwise f*cked up hands (lines blur together, they look like an elephant foot, etc) are the most common, but quite a few had inverted kneecaps (like a horse tbh), a backwards foot or feet that didn't touch the ground, a missing eye (not just the eye but the whole eye socket) and off-center browline, head too big in comparison to shoulders, there were actually a TON that were missing entire biceps (like Rayman type sh*t) (there were a couple that added scars to a character's nose who didn't have a scar there; it's unclear if it's a scar or an imperfection because Shad didn't zoom in on any of them, but imo adding scars makes anything cooler. The problem is the machine probably doesn't know it added a scar and if zoomed in on it will also probably look f*cked up)
Hell, one of them had a third hand attached to the back of the calf and the other had a third foot hanging from its cape
The problem is that people who aren't trained in anatomical drawing or art in general won't see these imperfections that, as an artist, are very visible, yet a good portion of them will still use them over actually hiring professional artists or actually learning how to draw themselves because a) they don't have time to put into learning a skill (which is valid), b) just don't want to put time into learning a skill (not so much), or c) don't want to pay actual artists for their work that is also already being sampled by these AI generators without the artist's permission anyway (look at what DeviantArt is doing)
And because they don't got a bitch in their ear (an actual professional artist), AI generated art is actually a masterpiece (horrors beyond human imagination)
And then they laugh at actual artists who a) protest the AI generators sampling their art without permission and/or b) will be impacted monetarily by AI art by basically saying sh*t like "git gud" when they don't even want to "git gud" themselves
Same sh*t happened when crypto bros started stealing professional artists' works and turning them into NFTs without their permission, including someone who already f*cking died like 2 years prior
It's not even a case of "democratizing art" (as if professional artists had some form of political power or otherwise any control over commissioners aside from declining commissions or [checks notes] being paid for a service), because "democratizing art" would be something like idk free art school for everyone who wants to take art classes
This is great so informative. I Love to see more videos like this. Plus some of the swords depicted in the images, just screamed mall ninja to me. 🤣🤣
This would make a great series.
I'm excited to see how the artwork for the All-Star series goes. Can't wait to read it!
What a time to be alive
11:55 the AI is proving that the stick is the ultimate weapon
Shad: "Let's go back to my design and give the AI a lot of freedom."
AI: generates endless Star Wars hookahs
Shad: "You're drunk."
Nah man, the program's not drunk, it's stoned.
Oh man I kinda wish imagine the possibilities if it was on shrooms it probably answer all of mankind’s questions of the universe.
That was a really fun video! I've been using AI image gen for a while and it never stops being, hilarious, impressive, and scary.
This "AI" seems to be a program that just composites multiple images together. You should be careful of using any art created by these programs because it may end up being art created by someone else that was just composited together by the program.
AI art generators are not compositing images, and Stable Diffusion 100% is not. They do take in lots of training data scraped from the internet, and there's a massive legal gray area around some of that right now, but you can be assured that no direct copy of any part of a training image is going to end up in your output.
I love this! I want to see you do more of this, perhaps try castles and then promptly JUDGE THEM for how terrible (or amazing) they'll look! Honestly, I made a castle for one of my attempts (literally my first attempt at a castle. An interior, no less) and it was FLAWLESS.
Every 60 seconds in Africa, a minute passes.
But together we can stop that...
Africa had its chance...
With your contribution of just $3 a month, we can get one step closer to reducing the temperature of the African continent to absolute 0, effective stopping time in Africa.
Shad's doing all the shit I want to do, been writing my own fantasy novels and have ideas for a space opera. Only known about him for two or three weeks now. Timing.
The AI art is constantly getting better. After you "train" the AI and some trial and error, AI can make some amazing pieces. Bad artists are big mad right now. The good artists have nothing to worry about. In fact many good artists are ok with AI, because it's a great tool.
Even an expert can draw *inspiration* from the madness the AI can make at times. There are a lot of "what the actual..." did the AI just do and you can really get inspired by it. Not just that but it can help you get out of your comfort zone in regards to art. It's a very handy tool, especially if you know what you're doing(ie cleaning up the mess etc).
It can and can't help. The impulse is to just let the AI do the work, which most people are doing and claiming they did it, not the AI. There's an ugly side to this and will dominate it pretty soon as well as the ethics around it. There's already a lawsuit with one site using it because of where it's pulling it's sources from, which might spread to the other AI programs. The temptation is make little edits and call it a day, because honestly a pro can see it did most the work unless they had a specific idea in mind. I don't see many of them using AI, but you do those that can't draw or not great.
Good artist aren't born fully formed from Zeus's forehead. They start out as bad artists. A generation with no market for bad art will mean none of them get enough commissions to stay in and become good artists. And all the existing good artists will die off
@@nathanbrown8680 Some are born talented, but a lot of them now lack imagination though keen understanding of technical skill and you see vice-versa. At least that is what I saw when with ex at her art school.
Bad artists will stop and give up or they will lean on AI to compete with the non-artists. There is already those people starting fights with artists and back n forth.
I don't think it will go as people think as A|I art has limitations and is not some supercomputer like people think; someone is programming it to do things. Most times I see it create similar illustrations or character pieces and becomes immediately identifiable. A non-artist can't do a commission with specific needs, but someone that can draw will and use the AI to enhance it.
Then there is the legal side of where AI is pulling its resources from. If companies jump into it, then you will see that go to court before individual artists and credit of some kind has to be given when using AI.
@@nathanbrown8680 What you just described is called competition. Even before AI, that competition is brutal. If the AI is weeding out all the subpar artists who refuse to improve, that's not the AI's fault. What I mean by bad artists are people like the current "comic book pros", the ones doing Calarts or drawing "gay Superman", etc. They're the biggest complainers about AI, because they know the AI will surpass them, if it hasn't already.
That's who I'm referring as bad artists. The artists who work on themselves and their craft, by default are always gonna be the better artist. And like I said, the good (and constantly improving) artists won't have to worry about AI.
You should try this with your castle designs as well. Would enjoy seeing a video of that too.
I'm more on the side of calling it "A.I. generated images".
Art is defined as "1. The conscious use of the imagination in the production of objects intended to be contemplated or appreciated as beautiful, as in the arrangement of forms, sounds, or words." and, as such, it requires imagination and inventive to fulfill an essential role in the creation of the image.
Making a description and adding examples to take from to do an image is not art, and neither is making a collage from the multiple results. Making that statement is calling a traced image "art".
In an example very close to you, Shad: imagine someone calling him/herself an author by taking "Shadows of the Conqueror" and changing 1 to 2 letters of the names (characters, places, things) and race/gender swapping characters and selling that book. Would you be pleased and celebrate him/her? Would you call him/her a fellow writer/author?
The bit about his writing is not at all like what is happening here, and the rest is a bit of a stretch for the analogy, as even a tracing can be artistic. Doing somebody else's piece in your style has been done many times before and is a common way for artists to learn new things or to practice, and the pieces that come out of it are still considered art. I trace over photos of actual landscapes and city skylines to make backgrounds and scenery shots for my illustrations, and what comes out of that process still meets the definitions of art.
While AI does do some of the work, as you saw here, the best results are from an already made concept image, which itself requires artistic skill to produce. The AI is another tool you can use to make the process easier. There are people who will pass off almost raw outputs of these programs as art, and that is disingenuous, but to call the end image that Shad ended up with here "not art" is a leap and a half. The amount of work and intentional design choices is still all there, just through a new and different process.
Great video as always Shad. You are a true inspiration to me.😉
In under one year, every person who works creating "weapons" and "armor" for fantasy games will be out of work, period.
Then again, it's their own fault. They created these abominations, now the AI simply picks those up and remixes them.
It may or may not be true but more unemployment is never a good thing.
this video is exactly what I have been doing just yesterday for around 2 hours
AI art is severly overrated. Most of it is utterly soulless and it is missing that human touch. And by most I mean all of it. People are dumb.
In my lowbrow uneducated understanding of the term "art", it is at it's best when used to accomplish something like conveying a message even, or just a feeling, a memory, an emotion. That can not be done by ai yet. A human artist with a vision will always be needed to form and orchestrate. You can feed the machine with association like algorithms and ideas and all, but it seems to be the turning point whether or not our creativity is more than just an idea. The machine can translate, but not innovate like we do.
human touches are overrated anyway especially my uncle's
I'm sorry for your loss.
to blend the blade into the coloration, you'll have to play around with the layer blending mode (and strength) of the blade (only the blade, not the hilt), and putting some glow from the blade on the character in additive blending mode.