| Hey people. I'm curious to hear some opinions on what you think of the current and future ability of AI art. More specifically when it comes to stable diffusion and the various models that exist and also the amount of advanced tools that exist that allow people to truly augment what they create [for example training dreambooth/lora models, using inpainting/outpainting, controlnet and all the other insane extensions that are constantly coming out]. As a computer science major and believer in technological progress I believe the end result of these technologies is as clear as day but I'm curious on what people here think. Exactly four days ago I decided to begin creating AI art using various diffusion models and also training models myself and overall the leap from what was possible in 2020 until now is truly staggering. Here's a full album of some examples of what I've created so far [total is 27 images out of most likely 900 generations that were trashed]. https://imgur.com/a/v0UI5fx Keep in mind due to the size limit of most platforms such as imgur [20mbs] or pixiv [32mbs] the true resolution of these images cannot be uploaded. Using gigapixel which is a paid industry grade AI upscaler these have been upscaled by 4x [making them 3994 x 3994 pixels each] however locally I have these at 31.25x upscale which come to a staggering 32000 x 32000 pixels and a size of 620mb each. Can you see this level of quality with current monitors? No, but it definitely makes the image far nicer and detailed than what is able to be uploaded online. Also if you want more art I upload more every day so feel free to check my pixiv out: https://www.pixiv.net/en/users/87340390 There's a total of 27 images in the album but here is a few to showcase. [font="\"Proxima Nova Regular\", \"Helvetica Neue\", Helvetica, Arial, sans-serif"]  [/font] [font="\"Proxima Nova Regular\", \"Helvetica Neue\", Helvetica, Arial, sans-serif"]  [/font] [font="\"Proxima Nova Regular\", \"Helvetica Neue\", Helvetica, Arial, sans-serif"]  [/font] [font="\"Proxima Nova Regular\", \"Helvetica Neue\", Helvetica, Arial, sans-serif"]  [/font][font="\"Proxima Nova Regular\", \"Helvetica Neue\", Helvetica, Arial, sans-serif"]  [/font][font="\"Proxima Nova Regular\", \"Helvetica Neue\", Helvetica, Arial, sans-serif"]  [/font] |
CneqFeb 22, 2023 2:47 AM
 |
| Art is a creative expression created by human skill and imagination. It can come in different visual ways, from sculpture to paintings to many other things. Art can also have a emotional power, and lives on among the people, even when the artist died. At that moment, the art becomes more precious because that's all we have left of the people that once lived. Art can also educate us, that's why I appreciated all kind of art, even art that tell us a dark and evil things. Can we also tell this about AI? Is it also a expression created by AI skill and imagination? ? AI doesn't understand emotions and empathy like biological lifeforms does, so does AI generated art hold the same emotional value as human made art? I'm also afraid that our lives are to much depended of technology. We can see it already in people's behaviour and how they got addicted to certain science and technology. Is it a good thing to make technology our primary way of living? Take for example GPS, nowadays everyone has GPS. Many years ago, you had to figure it out where to go, you always find a way, using your own skill and imagination to fix a issue from going A to B without knowing the way. Nowadays you type in a place, and it takes you there. The problem with this is, you create a tunnel vision, it takes over some of your "skills" and at that moment, you start forgetting how to solve a problem. And at that moment when someone takes away your GPS, you don't know it anymore, most people would panic. Just simply by the fact that you can't find your own way anymore, because the GPS was always hand holding you. Now let us reflect his towards AI art. If AI art is going to take over, than we start to forget how to make precious art on our own, we start to lose our own skill and imagination, which we need more than ever before. I'm not completely against technology and sience, I also love it. But I can see the dark side aswell. We should preserve our own consciousness, skill, imagination, free mind and above all humanity. Nothing should take control over that, not even a AI. I'm a proponent for technology regulations, including AI. It should be a help tool, not become our primary way of living. And it should never take over our own skill and imaginations, we should preserve that. |
Toonen1988Feb 25, 2023 4:54 PM
| "Most people talk about killing time while time is killing them. You can outrun everything but you'll never outrun the hands of time. Use it wisely before you expire". - Toonen1988 "Cyberpunk show us the dark side, reveiling the dangerous side effects of the drug of futurism." - Indigo Gaming |
Toonen1988 said: I'd suggest you actually try using stable diffusion and learn how it even works because it's clear you have no idea what "AI Art" even is.Art is a creative expression created by human skill and imagination. It can come in different visual ways, from sculpture to paintings to many other things. Art can also have a emotional power, and lives on among the people, even when the artist died. At that moment, the art becomes more precious because that's all we have left of the people that once lived. Art can also educate us, that's why I appreciated all kind of art, even art that tell us a dark and evil things. Can we also tell this about AI? Is it also a expression created by AI skill and imagination? ? AI doesn't understand emotions and empathy like biological lifeforms does, so does AI generated art hold the same emotional value as human made art? I'm also afraid that our lives are to much depended of technology. We can see it already in people's behaviour and how they got addicted to certain science and technology. Is it a good thing to make technology our primary way of living? Take for example GPS, nowadays everyone has GPS. Many years ago, you had to figure it out where to go, you always find a way, using your own skill and imagination to fix a issue from going A to B without knowing the way. Nowadays you type in a place, and it takes you there. The problem with this is, you create a tunnel vision, it takes over some of your "skills" and at that moment, you start forgetting how to solve a problem. And at that moment when someone takes away your GPS, you don't know it anymore, most people would panic. Just simply by the fact that you can't find your own way anymore, because the GPS was always hand holding you. Now let us reflect his towards AI art. If AI art is going to take over, than we start to forget how to make precious art on our own, we start to lose our own skill and imagination, which we need more than ever before. I'm not completely against technology and sience, I also love it. But I can see the dark side aswell. We should preserve our own consciousness, skill, imagination, free mind and above all humanity. Nothing should take control over that, not even a AI. I'm a proponent for technology regulations, including AI. It should be a help tool, not become our primary way of living. And it should never take over our own skill and imaginations, we should preserve that. AI art is not "generated" out of thin air, everything is specified by very specific prompts to the most granular detail you can imagine and that is just the beginning. First off everything is powered first by a general diffusion model and many hundreds of them exist. These core models can be trained yourself and/or you can fuse them to create entirely new models. These core models are extremely diverse in what they have been trained on and what they provide and every single one of them has a use. There also exists LORA/Dreambooth models/concepts can range from anything, an item, a person, an abstract idea or an art style you name it. These "concepts" can be used in conjunction with one another and combining these concepts [which are all trained by people and can be trained by yourself, which is what I do] are then used to generate a first attempt which uses extreme computational power. These power can range depending on what you are making however in the case of high quality art you need to natively upscale a 512x512 by 2x and this requires at minimum 12gb+ of VRAM and even then you will see crashes. 16gb+ of VRAM is required to generate any art that can reasonably be seen as high quality. After this begins for every one "good" generation of the same set of prompts/concepts/embeddings you will have 100 "bad" generations that range from anything from mutated limbs, bad hands and other anomalies that are a limitation of the current process of stable diffusion. These can be offset by using another set of "negative prompts" and this is also mandatory on top of using your other prompts as well as concepts/textual embeddings. After you find a suitable generation out of 50+ to 100 questionable generations the next step is to use a process called inpainting and a process called outpainting to enhance/change anything about this generation to make it fit more to your vision. This is a process of cropping out certain aspects of the generation in img2img and to regenerate certain aspects again or change things entirely. After this is finished due to a 2x of 512x512 being extremely low quality [despite requiring a $7000+ PC setup] it is then a good idea to use a secondary AI upscaler such as gigapixel which can scale images up to 31.25x their initial resolution and for most websites at max they can support 6x. This software is expensive and there does not exist any free alternatives that come anywhere near the quality if you do not have the computational power to natively upscale on generation beyond 2x. And this is just for very, very basic generations. There now exists modules such as control net and many others that are constantly releasing only furthering added to how involved actually creating AI art actually is. Anyone who is competent with this technology is involved 100% of the time and I will say right now it takes far more technical ability to manually train concepts and apply all these moving parts to create one unified vision and even in it's current state it is 100% comparable to a digital artist using photoshop [inpainting/outpainting is literally just an advanced version of what photoshop wished it could be]. AI art is nothing more than the next evolution of digital art and it is a tool to be used by a creator in the same way digital tools such as photoshop were used. Photoshop significantly cut down on what amount of manual work was required of an artist to create art and when it comes to stable diffusion this is simply the next evolution of that. The more mundane monotonous labour of art is once again being shuttered away to allow people to focus more on their creativity and imagination. |
|
Toonen1988 said: Art is a creative expression created by human skill and imagination. It can come in different visual ways, from sculpture to paintings to many other things. Art can also have a emotional power, and lives on among the people, even when the artist died. At that moment, the art becomes more precious because that's all we have left of the people that once lived. Art can also educate us, that's why I appreciated all kind of art, even art that tell us a dark and evil things. Can we also tell this about AI? Is it also a expression created by AI skill and imagination? ? AI doesn't understand emotions and empathy like biological lifeforms does, so does AI generated art hold the same emotional value as human made art? I'm also afraid that our lives are to much depended of technology. We can see it already in people's behaviour and how they got addicted to certain science and technology. Is it a good thing to make technology our primary way of living? Take for example GPS, nowadays everyone has GPS. Many years ago, you had to figure it out where to go, you always find a way, using your own skill and imagination to fix a issue from going A to B without knowing the way. Nowadays you type in a place, and it takes you there. The problem with this is, you create a tunnel vision, it takes over some of your "skills" and at that moment, you start forgetting how to solve a problem. And at that moment when someone takes away your GPS, you don't know it anymore, most people would panic. Just simply by the fact that you can't find your own way anymore, because the GPS was always hand holding you. Now let us reflect his towards AI art. If AI art is going to take over, than we start to forget how to make precious art on our own, we start to lose our own skill and imagination, which we need more than ever before. I'm not completely against technology and sience, I also love it. But I can see the dark side aswell. We should preserve our own consciousness, skill, imagination, free mind and above all humanity. Nothing should take control over that, not even a AI. I'm a proponent for technology regulations, including AI. It should be a help tool, not become our primary way of living. And it should never take over our own skill and imaginations, we should preserve that. I agree with this. I also have major qualms about the training data used for these AI generators - they don't exist in a vacuum, they were trained on existing artwork, work people poured years of learning and their hearts and souls into. Those images were scraped from the internet and compiled into a dataset without the original artists' knowledge or consent, and then used to train image generators, to show them the patterns people gravitate to when creating art so that it could replicate those patterns. That's all AI "art" is, it's pattern replication, it can't innovate or truly create, and the fact that it was trained on peoples' hard work without so much as a heads up is revolting to me. As an artist whose work is in those datasets, it feels like a slap in the face to be told that AI generators are "the next evolution of digital art" - they're not, they can't evolve. They can't do anything new. Comparing them to programs like photoshop wherein artists literally paint their works is just laughable, and only shows me that OP knows very little about the creation of digital art. At best, AI generators could be useful for thumbnailing, but not the way they currently stand, built on stolen work. The datasets should be comprised of public domain works, and works that were voluntarily opted in by the original creator, where the creator was compensated for the use of their work. Until the datasets are sourced ethically, I cannot view AI image generators with anything but disgust. Lastly, there is the issue of oversaturation posed by AI content, which I think this tweet sums up nicely. (translation) I believe that the grave situation brought about by the spread of AI illustrations is not a labour problem but an environmental one. Since AI content can be produced more quickly that hand-drawn art, it will not be very long before the number of AI-generated art exceeds the total number of regular art and photos. Once the internet is overflowing with AI content, image searches will become unusable. The introduction of a non-native species with strong reproductive ability and the resultant loss of diversity within the ecosystem is not something that can be called progress. |
| I can generate thousands of images every day of my obscure waifu at a level consistently higher than a majority of artists, even on my shitty 6gb VRAM card. The output is no less creative than most human art I've seen and largely limited by my own prompting. Anything it struggles with you can teach it in a couple hours. I've probably spent more time autistically fine-tuning LoRAs than actually prompting lately. I'm also not convinced that it's somehow 'unethical' - real artists learn by studying the real world and other artists right? I don't see how that's fundamentally much different from how you teach machines to do it, myself. |
 |
Cneq said: ... if you want more art I upload more every day so feel free to check my pixiv out: https://www.pixiv.net/en/users/87340390 your art is wonderful in it's sense of delicate, gentle beauty and cuteness. Congrats~ |
 |
| AI art and photos can look good but there's something missing and uncanny with literally every single pic. Maybe some people can't see the uncanniness I'm referring to for whatever reason, but I bet a lot of people do notice. I know there's going to be a 1 out of a million pic that will fool me which someone can post here, but I imagine that when I learn a picture that "fooled me" is actually AI, I would see the uncanniness in it from that point on. Besides the obvious mistakes it makes sometimes, there's something like a soul/heart in real art which AI can't seem to copy yet. It's an unexplainable factor missing from their work. People think there is no such thing as soul/energy/heart/spirits etc but that is retarded naïve thinking; these things are very real even when they can't be measured. We are so used to it we take it for granted. For example, I think if today's AI built cities we would notice the difference from human cities quickly. It may take a human brain to understand that missing factor, or maybe AI can be coached enough to pass. The latter would be cool! All that said I think the AI art will be cool for building websites and stuff. Instead of original designs, maybe it would be better at remixing already-made art assets though it sounds like it already sort of does that. AI in customer service and coding will be very strong. btw I didn't read any of the text here, and I skimmed over the art as it's kinda bright in a way that's uncomfortable to view so I didn't look closely. So if I missed something, sorry! |
Shishio-kunFeb 26, 2023 10:37 AM
Toonen1988 said: Digital art is just pixels on your screen. If there's an AI smart enough to find the right assortment of pixels to a degree where you can't tell the difference anymore, then clearly none of these things like "understanding of emotions, empathy etc." are required for the process of making art.Art is a creative expression created by human skill and imagination. It can come in different visual ways, from sculpture to paintings to many other things. Art can also have a emotional power, and lives on among the people, even when the artist died. At that moment, the art becomes more precious because that's all we have left of the people that once lived. Art can also educate us, that's why I appreciated all kind of art, even art that tell us a dark and evil things. Can we also tell this about AI? Is it also a expression created by AI skill and imagination? ? AI doesn't understand emotions and empathy like biological lifeforms does, so does AI generated art hold the same emotional value as human made art? I know it sucks to admit that we're losing yet another domain to the machines, but it's not like this is the first. This is no different than admitting that the machine is better at playing chess than us. There are a few things we have left, but art is no longer one of these. If you disagree, just wait another year or two. It'll become undeniable eventually. @cneq nice art, seriously. Looking very clean. |
| *lampoons inwardly* |
Cneq said: Toonen1988 said: I'd suggest you actually try using stable diffusion and learn how it even works because it's clear you have no idea what "AI Art" even is.Art is a creative expression created by human skill and imagination. It can come in different visual ways, from sculpture to paintings to many other things. Art can also have a emotional power, and lives on among the people, even when the artist died. At that moment, the art becomes more precious because that's all we have left of the people that once lived. Art can also educate us, that's why I appreciated all kind of art, even art that tell us a dark and evil things. Can we also tell this about AI? Is it also a expression created by AI skill and imagination? ? AI doesn't understand emotions and empathy like biological lifeforms does, so does AI generated art hold the same emotional value as human made art? I'm also afraid that our lives are to much depended of technology. We can see it already in people's behaviour and how they got addicted to certain science and technology. Is it a good thing to make technology our primary way of living? Take for example GPS, nowadays everyone has GPS. Many years ago, you had to figure it out where to go, you always find a way, using your own skill and imagination to fix a issue from going A to B without knowing the way. Nowadays you type in a place, and it takes you there. The problem with this is, you create a tunnel vision, it takes over some of your "skills" and at that moment, you start forgetting how to solve a problem. And at that moment when someone takes away your GPS, you don't know it anymore, most people would panic. Just simply by the fact that you can't find your own way anymore, because the GPS was always hand holding you. Now let us reflect his towards AI art. If AI art is going to take over, than we start to forget how to make precious art on our own, we start to lose our own skill and imagination, which we need more than ever before. I'm not completely against technology and sience, I also love it. But I can see the dark side aswell. We should preserve our own consciousness, skill, imagination, free mind and above all humanity. Nothing should take control over that, not even a AI. I'm a proponent for technology regulations, including AI. It should be a help tool, not become our primary way of living. And it should never take over our own skill and imaginations, we should preserve that. AI art is not "generated" out of thin air, everything is specified by very specific prompts to the most granular detail you can imagine and that is just the beginning. First off everything is powered first by a general diffusion model and many hundreds of them exist. These core models can be trained yourself and/or you can fuse them to create entirely new models. These core models are extremely diverse in what they have been trained on and what they provide and every single one of them has a use. There also exists LORA/Dreambooth models/concepts can range from anything, an item, a person, an abstract idea or an art style you name it. These "concepts" can be used in conjunction with one another and combining these concepts [which are all trained by people and can be trained by yourself, which is what I do] are then used to generate a first attempt which uses extreme computational power. These power can range depending on what you are making however in the case of high quality art you need to natively upscale a 512x512 by 2x and this requires at minimum 12gb+ of VRAM and even then you will see crashes. 16gb+ of VRAM is required to generate any art that can reasonably be seen as high quality. After this begins for every one "good" generation of the same set of prompts/concepts/embeddings you will have 100 "bad" generations that range from anything from mutated limbs, bad hands and other anomalies that are a limitation of the current process of stable diffusion. These can be offset by using another set of "negative prompts" and this is also mandatory on top of using your other prompts as well as concepts/textual embeddings. After you find a suitable generation out of 50+ to 100 questionable generations the next step is to use a process called inpainting and a process called outpainting to enhance/change anything about this generation to make it fit more to your vision. This is a process of cropping out certain aspects of the generation in img2img and to regenerate certain aspects again or change things entirely. After this is finished due to a 2x of 512x512 being extremely low quality [despite requiring a $7000+ PC setup] it is then a good idea to use a secondary AI upscaler such as gigapixel which can scale images up to 31.25x their initial resolution and for most websites at max they can support 6x. This software is expensive and there does not exist any free alternatives that come anywhere near the quality if you do not have the computational power to natively upscale on generation beyond 2x. And this is just for very, very basic generations. There now exists modules such as control net and many others that are constantly releasing only furthering added to how involved actually creating AI art actually is. Anyone who is competent with this technology is involved 100% of the time and I will say right now it takes far more technical ability to manually train concepts and apply all these moving parts to create one unified vision and even in it's current state it is 100% comparable to a digital artist using photoshop [inpainting/outpainting is literally just an advanced version of what photoshop wished it could be]. AI art is nothing more than the next evolution of digital art and it is a tool to be used by a creator in the same way digital tools such as photoshop were used. Photoshop significantly cut down on what amount of manual work was required of an artist to create art and when it comes to stable diffusion this is simply the next evolution of that. The more mundane monotonous labour of art is once again being shuttered away to allow people to focus more on their creativity and imagination. Afcourse, every AI needs to be trained, otherwise the AI doesn't know what a astronaut or a horse is. So, how can it generate a astronaut on a horse? This is just simple logic that every simple mind can understand. But as soon the AI knows what a astronaut and a horse is, then you can generate a image with a astronaut on a horse. But this doesn't change the fact that you don't need ANY skills anymore to draw a picture yourself. So therefor, how more people are going to use this kind of technology, how less and less people can make art on their own, and with their hands. You don't need any skill anymore to create "art" and everyone can do it. It's not special anymore, you don't need to have any art skills to make something. When the AI is trained, you give indeed some promps and it generate a picture. It doesn't make you a artist, you don't need any skills or what so ever. I followed the training for stable diffusion and you don't need to have any artist skill. You can also just download .ckpt files and there you go, you can create even more art by just giving in some prompts. Soon or later, people don't even know what a pencil is anymore. Same with music, nowadays you can also generate music. Just give in some prompts, it generates. What are you going to say afterwards? "Look what I got, I made a song." No, you don't, the AI did. You only give in some prompts. You are nowhere compared to people who learned instruments for years. And it doesn't matter who trained it, it still doesn't make you a music producer or instrument player. Professional instrument players dedicated their lives to it. You won't be near any of their level with a AI. This goes for other art aswell, including the art we are talking about. |
Toonen1988Feb 26, 2023 2:38 PM
| "Most people talk about killing time while time is killing them. You can outrun everything but you'll never outrun the hands of time. Use it wisely before you expire". - Toonen1988 "Cyberpunk show us the dark side, reveiling the dangerous side effects of the drug of futurism." - Indigo Gaming |
syrcus said: Toonen1988 said: Art is a creative expression created by human skill and imagination. It can come in different visual ways, from sculpture to paintings to many other things. Art can also have a emotional power, and lives on among the people, even when the artist died. At that moment, the art becomes more precious because that's all we have left of the people that once lived. Art can also educate us, that's why I appreciated all kind of art, even art that tell us a dark and evil things. Can we also tell this about AI? Is it also a expression created by AI skill and imagination? ? AI doesn't understand emotions and empathy like biological lifeforms does, so does AI generated art hold the same emotional value as human made art? I'm also afraid that our lives are to much depended of technology. We can see it already in people's behaviour and how they got addicted to certain science and technology. Is it a good thing to make technology our primary way of living? Take for example GPS, nowadays everyone has GPS. Many years ago, you had to figure it out where to go, you always find a way, using your own skill and imagination to fix a issue from going A to B without knowing the way. Nowadays you type in a place, and it takes you there. The problem with this is, you create a tunnel vision, it takes over some of your "skills" and at that moment, you start forgetting how to solve a problem. And at that moment when someone takes away your GPS, you don't know it anymore, most people would panic. Just simply by the fact that you can't find your own way anymore, because the GPS was always hand holding you. Now let us reflect his towards AI art. If AI art is going to take over, than we start to forget how to make precious art on our own, we start to lose our own skill and imagination, which we need more than ever before. I'm not completely against technology and sience, I also love it. But I can see the dark side aswell. We should preserve our own consciousness, skill, imagination, free mind and above all humanity. Nothing should take control over that, not even a AI. I'm a proponent for technology regulations, including AI. It should be a help tool, not become our primary way of living. And it should never take over our own skill and imaginations, we should preserve that. I agree with this. I also have major qualms about the training data used for these AI generators - they don't exist in a vacuum, they were trained on existing artwork, work people poured years of learning and their hearts and souls into. Those images were scraped from the internet and compiled into a dataset without the original artists' knowledge or consent, and then used to train image generators, to show them the patterns people gravitate to when creating art so that it could replicate those patterns. That's all AI "art" is, it's pattern replication, it can't innovate or truly create, and the fact that it was trained on peoples' hard work without so much as a heads up is revolting to me. As an artist whose work is in those datasets, it feels like a slap in the face to be told that AI generators are "the next evolution of digital art" - they're not, they can't evolve. They can't do anything new. Comparing them to programs like photoshop wherein artists literally paint their works is just laughable, and only shows me that OP knows very little about the creation of digital art. At best, AI generators could be useful for thumbnailing, but not the way they currently stand, built on stolen work. The datasets should be comprised of public domain works, and works that were voluntarily opted in by the original creator, where the creator was compensated for the use of their work. Until the datasets are sourced ethically, I cannot view AI image generators with anything but disgust. Lastly, there is the issue of oversaturation posed by AI content, which I think this tweet sums up nicely. (translation) I believe that the grave situation brought about by the spread of AI illustrations is not a labour problem but an environmental one. Since AI content can be produced more quickly that hand-drawn art, it will not be very long before the number of AI-generated art exceeds the total number of regular art and photos. Once the internet is overflowing with AI content, image searches will become unusable. The introduction of a non-native species with strong reproductive ability and the resultant loss of diversity within the ecosystem is not something that can be called progress. Thank you for explaining it a bit more in detail. |
| "Most people talk about killing time while time is killing them. You can outrun everything but you'll never outrun the hands of time. Use it wisely before you expire". - Toonen1988 "Cyberpunk show us the dark side, reveiling the dangerous side effects of the drug of futurism." - Indigo Gaming |
Toonen1988 said: You seem completely oblivious that prompts do not create the majority of AI art and almost anyone competent relies on inpainting/out painting and numerous combinations of LORA concepts/textual embeddings. Cneq said: I'd suggest you actually try using stable diffusion and learn how it even works because it's clear you have no idea what "AI Art" even is. AI art is not "generated" out of thin air, everything is specified by very specific prompts to the most granular detail you can imagine and that is just the beginning. First off everything is powered first by a general diffusion model and many hundreds of them exist. These core models can be trained yourself and/or you can fuse them to create entirely new models. These core models are extremely diverse in what they have been trained on and what they provide and every single one of them has a use. There also exists LORA/Dreambooth models/concepts can range from anything, an item, a person, an abstract idea or an art style you name it. These "concepts" can be used in conjunction with one another and combining these concepts [which are all trained by people and can be trained by yourself, which is what I do] are then used to generate a first attempt which uses extreme computational power. These power can range depending on what you are making however in the case of high quality art you need to natively upscale a 512x512 by 2x and this requires at minimum 12gb+ of VRAM and even then you will see crashes. 16gb+ of VRAM is required to generate any art that can reasonably be seen as high quality. After this begins for every one "good" generation of the same set of prompts/concepts/embeddings you will have 100 "bad" generations that range from anything from mutated limbs, bad hands and other anomalies that are a limitation of the current process of stable diffusion. These can be offset by using another set of "negative prompts" and this is also mandatory on top of using your other prompts as well as concepts/textual embeddings. After you find a suitable generation out of 50+ to 100 questionable generations the next step is to use a process called inpainting and a process called outpainting to enhance/change anything about this generation to make it fit more to your vision. This is a process of cropping out certain aspects of the generation in img2img and to regenerate certain aspects again or change things entirely. After this is finished due to a 2x of 512x512 being extremely low quality [despite requiring a $7000+ PC setup] it is then a good idea to use a secondary AI upscaler such as gigapixel which can scale images up to 31.25x their initial resolution and for most websites at max they can support 6x. This software is expensive and there does not exist any free alternatives that come anywhere near the quality if you do not have the computational power to natively upscale on generation beyond 2x. And this is just for very, very basic generations. There now exists modules such as control net and many others that are constantly releasing only furthering added to how involved actually creating AI art actually is. Anyone who is competent with this technology is involved 100% of the time and I will say right now it takes far more technical ability to manually train concepts and apply all these moving parts to create one unified vision and even in it's current state it is 100% comparable to a digital artist using photoshop [inpainting/outpainting is literally just an advanced version of what photoshop wished it could be]. AI art is nothing more than the next evolution of digital art and it is a tool to be used by a creator in the same way digital tools such as photoshop were used. Photoshop significantly cut down on what amount of manual work was required of an artist to create art and when it comes to stable diffusion this is simply the next evolution of that. The more mundane monotonous labour of art is once again being shuttered away to allow people to focus more on their creativity and imagination. Afcourse, every AI needs to be trained, otherwise the AI doesn't know what a astronaut or a horse is. So, how can it generate a astronaut on a horse? This is just simple logic that every simple mind can understand. But as soon the AI knows what a astronaut and a horse is, then you can generate a image with a astronaut on a horse. But this doesn't change the fact that you don't need ANY skills anymore to draw a picture yourself. So therefor, how more people are going to use this kind of technology, how less and less people can make art on their own, and with their hands. You don't need any skill anymore to create "art" and everyone can do it. It's not special anymore, you don't need to have any art skills to make something. When the AI is trained, you give indeed some promps and it generate a picture. It doesn't make you a artist, you don't need any skills or what so ever. I followed the training for stable diffusion and you don't need to have any artist skill. You can also just download .ckpt files and there you go, you can create even more art by just giving in some prompts. Soon or later, people don't even know what a pencil is anymore. Same with music, nowadays you can also generate music. Just give in some prompts, it generates. What are you going to say afterwards? "Look what I got, I made a song." No, you don't, the AI did. You only give in some prompts. You are nowhere compared to people who learned instruments for years. And it doesn't matter who trained it, it still doesn't make you a music producer or instrument player. Professional instrument players dedicated their lives to it. You won't be near any of their level with a AI. This goes for other art aswell, including the art we are talking about. Not to mention control net LITERALLY allows you to use a tablet to draw and control net just came out very recently. Every single week more and more power is given to stable diffusion and that power is the ability to integrate humans into the workflow and by extension making the already powerful generation tools far more powerful and quite frankly eliminating most issues that AI art has [such as drawing limbs and other anatomical issues]. Also "you don't need any art skills or what so ever." what is defined by "art skill" ? Art at it's foundation is built upon creativity and imagination and taking the formless and giving it form. How you accomplish this has always been changing and with the advent of digital art more and more of the manual process has been done away and many artists are entirely digital nowadays. AI is simply a radical acceleration of this trend and much like it has done for most industries it is extremely disruptive and essentially a leap no one was expecting. Putting aside the fact you can literally draw with AI art [and the fact you don't even realize this exists] the very idea that "art" is entirely defined by the tool you use and not by the artistic expression of art and how aesthetically pleasing it is is nothing more than cope. I can't see any logical basis behind this other than a stubborn refusal to cling to the past. Have you even browsed the hundreds of image boards and art sites and seen what "true" artists often upload? The 60% of good traditional artists are of high quality but the more mediocre artists have just as many issues when it comes to anatomically correct art that AI currently suffer from and even then AI art has far more character and diversity than what your average mediocre artist can achieve. The very notion that something is "off" about AI art or that it doesn't have a soul is due to the amount of soulless AI art that has been spewed across the internet with most using diffusion models BEFORE mixing became the norm [with mixing just now starting you have masterpiece level models such as orangemix] and this change has only happened very recently. Not to mention most people do not have the computational power of a 3080ti+ and something like gigapixel to get the best out of these models even if they did have the ability to and thus this further isolates the amount of quality AI art that exists online. And then you have midjourney and NovelAI [which the majority of people use that can't get into stable diffusion] and while these are "good" they do not have the same level of complexity and functionality of using stable diffusion with model mixes, lora concepts, control net etc and thus you once again see far more "soulless" AI art purely due to where and what type of people are making them. If you're going to say AI art is "soulless" despite not even knowing the reason for this [and why the reason essentially makes such a claim irrelevant] you should instead actually research the chronological path of going from GANs in 2017 all the way to advanced stable diffusion today [which in itself has only just begun] and you would realize we are most likely 1-3 years away from AI art being completely rid of most issues of "soullessness" purely by a result of more control being given to the user [and by extension morphing these tools into a super advanced photoshop]. Until mass adoption of more sophisticated AI tools becomes the norm you will still see "soulless" AI art but that has nothing to do with AI art itself and the innovation of these tools have essentially made such a concern almost irrelevant. For mine on the other hand the first time I ever touched AI art was on the 18th of February [little over a week ago] and prior to this I did not use NovelAI or any other platforms. With this being the case I can see some issues with my art when it comes to eye position and some other minor grievances. With more use of inpainting and control net with my tablet these technologies have the ability give power to people who are creative and imaginative yet do not have the time to manually draw. Likewise to @syrcus " That's all AI "art" is, it's pattern replication, it can't innovate or truly create, and the fact that it was trained on peoples' hard work without so much as a heads up is revolting to me." This is incredibly hilarious and it's almost as if you're still living in 2017 when GAN's were the extent of AI generation. Even with stable diffusion in it's relative infantile state it literally has the ability to utilize all art/styles/concepts/ideas and to create/innovate on concepts that LITERALLY haven't been combined before and to do so to such a degree that these fusions of styles/concepts most likely would've never existed due to how niche they may be and how limited artists to fit these niches simply do not exist and/or quit art or do not have the skill to do so. If you are going to argue that style replication to breed new styles is "not truly creating" and "not innovating" you'd have to say any artist who takes inspiration from other artists/styles and concepts/themes is "not truly creating" which is a bunch of bullshit. Overall this entire outcry against AI art seems to be entirely the result of the stereotypical response to a disruptive technology and by extension people being angry [rightfully so] that skills they had and once were valued have seen a significant drop in a very short period of time. Who is to blame for this? No one. Technological progress does not care for anyone and regardless if people want it or not. It is here, it is profoundly useful and it does not care for the human aspect of devaluing human skill. |
|
Cneq said: I think you're half right, but I think people have legitimate concerns about their own work being used for things that can be monetized without their consent. Yes, nobody likes feeling made obsolete and that's certainly why a lot of people are threatened by this stuff, but there are serious legal concerns even by people who are invested in its success. Overall this entire outcry against AI art seems to be entirely the result of the stereotypical response to a disruptive technology and by extension people being angry [rightfully so] that skills they had and once were valued have seen a significant drop in a very short period of time. Who is to blame for this? No one. Technological progress does not care for anyone and regardless if people want it or not. It is here, it is profoundly useful and it does not care for the human aspect of devaluing human skill. I think AI art is pretty interesting, and I think the future will see artists using AI as a baseline for inspiration, prototyping ideas, and maybe it improves their workflow. Still a long way to go though; I think the images you showed are catching at a glance but they all have serious problems. The hands, gun, and pants in the first image, for instance, are pretty horrendous and the shadows need work. Hands in general remain the defining feature of when you can tell something is AI generated, but shadows, body proportions, etc. are also routine problems that all AI art still don't really quite understand. I think this is actually a pretty solid instance where maybe an artist might generate an image like this and then touch up the details. |
| If it can make hentai of my fav waifus than I am all for our ai overlords. |
 |
Yudina said: The very concern that the way these models are trained somehow equates to "theft" is in itself the very issue. Most artists do not have a deep technical background and thus have absolutely no idea how these models are actually trained and in reality they do not "steal" anything but see existing art, look for what is "correct" and/or what constitutes a concept [or character] and after they have "learned" what make up these things they generate something entirely new and unique that does not exist.Cneq said: I think you're half right, but I think people have legitimate concerns about their own work being used for things that can be monetized without their consent. Yes, nobody likes feeling made obsolete and that's certainly why a lot of people are threatened by this stuff, but there are serious legal concerns even by people who are invested in its success. Overall this entire outcry against AI art seems to be entirely the result of the stereotypical response to a disruptive technology and by extension people being angry [rightfully so] that skills they had and once were valued have seen a significant drop in a very short period of time. Who is to blame for this? No one. Technological progress does not care for anyone and regardless if people want it or not. It is here, it is profoundly useful and it does not care for the human aspect of devaluing human skill. I think AI art is pretty interesting, and I think the future will see artists using AI as a baseline for inspiration, prototyping ideas, and maybe it improves their workflow. Still a long way to go though; I think the images you showed are catching at a glance but they all have serious problems. The hands, gun, and pants in the first image, for instance, are pretty horrendous and the shadows need work. Hands in general remain the defining feature of when you can tell something is AI generated, but shadows, body proportions, etc. are also routine problems that all AI art still don't really quite understand. I think this is actually a pretty solid instance where maybe an artist might generate an image like this and then touch up the details. Would you say an artist watching an anime and then using that art style as a reference when creating that character is "theft" because they are essentially copying what makes up that character and turning it into a different work? Of course not. Someone making one piece fan art and trying to mimick eiichiro oda style based on what can be seen and what is common in his style is almost exactly what current stable diffusion does [although in a very odd way]. There exists this completely naive and absurd misconception that AI art is nothing more than taking bits and pieces of existing art [as they are] and just slapping them together to make something new which is complete bullshit and is not the case with stable diffusion. In terms legality it's not like there is some active web scraper that grabs all art the moment it is uploaded. The majority of art [such as low quality art] is never touched since it's bad for training models and likewise most of this art is not copyrighted in the slightest and exists [and has existed] in the public domain for a very long time. The distinction of how these models are trained which is based on the AI learning the concept itself [and not simply ripping the existing art and slapping it together with other concepts] is key and most legal action is not going to come as a result of some "copyright" bullshit but instead due to how harmful this type of generative technology can be in general [more specifically the fact it's a more faster and easier way to use deep fakes and this could be used to spread massive misinformation and all other issues]. That will be the reason for any legal action against it and by no means a result of it some being unethical to "steal" art since it's not stealing it in the first place. Although I wouldn't be surprised if they used it as a moral high ground talking point to spew to the general public if they did try to push to ban all generative art [which may I add is impossible since these models already exist and are not reliant on any third party besides your computer hardware so I doubt they will try to ban them regardless]. Also as I've said the current issues with hands and anything more complex [like guns] are a temporary setback that all AI generative works cannot cope with and besides inpainting these things over and over again to achieve a desired result it will take time [which I suspect will be another 1-3 years]. When it comes to lighting and shadows before orangemix and other new "mixes" lighting was a serious issue but even with these mixes being relatively new [and more being made all the time] this issue is almost gone even with basic prompting and zero inpainting/outpainting or control net tweaks. Such as in this case: https://www.pixiv.net/en/artworks/105726514 I could find hundreds of examples of bad lighting/shadows from amateur traditional artists and yet most people using orangemix or something similar the lighting far and wide is superior. Will it be superior to top tier artists in this state? No. However what number of top tier artists exist in the current population? and how many of those people are available to cater to specific niches? characters and ideas? Hardly any. Thus in the case AI generated art undoubtedly still wins since it can create art for more niche characters/themes [which are fundamentally lacking] and likewise is superior to most art that 40% of people make. Purely on the first page alone for this character most traditional art is simply lacking: https://www.pixiv.net/en/artworks/105428168 https://www.pixiv.net/en/artworks/105148658 https://www.pixiv.net/en/artworks/105191762 https://www.pixiv.net/en/artworks/105131035 https://www.pixiv.net/en/artworks/105132253 I would personally rather view AI generated art compared to these and I would struggle to see why anyone would think otherwise lol |
|
Cneq said: I don't actually agree that a lot of AI generated art is necessarily superior. A lot of what you showed me for instance, specifically the ones with guns or weapons, actually remind me of a lot of early 2000s art that I would find on pixiv / danbooru, and I actually find that the way they're drawn is really derivative and uninspired in comparison. Now, granted, there's an argument to be made that technically speaking a lot of AI art shows a lot of promise and yeah a ton of artists can't stack up against this stuff, but part of me also just thinks what makes art great is in the purpose that we can derive out of why someone made a particular stroke or color selection.Also as I've said the current issues with hands and anything more complex [like guns] are a temporary setback that all AI generative works cannot cope with and besides inpainting these things over and over again to achieve a desired result it will take time [which I suspect will be another 1-3 years]. When it comes to lighting and shadows before orangemix and other new "mixes" lighting was a serious issue but even with these mixes being relatively new [and more being made all the time] this issue is almost gone even with basic prompting and zero inpainting/outpainting or control net tweaks. Such as in this case: https://www.pixiv.net/en/artworks/105726514 I could find hundreds of examples of bad lighting/shadows from amateur traditional artists and yet most people using orangemix or something similar the lighting far and wide is superior. Will it be superior to top tier artists in this state? No. However what number of top tier artists exist in the current population? and how many of those people are available to cater to specific niches? characters and ideas? Hardly any. Thus in the case AI generated art undoubtedly still wins since it can create art for more niche characters/themes [which are fundamentally lacking] and likewise is superior to most art that 40% of people make. Purely on the first page alone for this character most traditional art is simply lacking: https://www.pixiv.net/en/artworks/105428168 https://www.pixiv.net/en/artworks/105148658 https://www.pixiv.net/en/artworks/105191762 https://www.pixiv.net/en/artworks/105131035 https://www.pixiv.net/en/artworks/105132253 I would personally rather view AI generated art compared to these and I would struggle to see why anyone would think otherwise lol So to take the additional picture you posted, the reason why the AI decided to draw light above the character's nose when it makes no sense from a shadow perspective is fascinating as a technological question, but it doesn't interest me from an artist or aesthetic perspective. There is, at least not in the moment, no aesthetic question that I can ask the AI that will return a comprehensible answer with respect to art, technique, and method. That would be up to the artist who is creating the AI art and that could be an interesting discussion. And again, I think that's going to be the interesting future, what do artists do with AI and appropriate it for their needs and vision. On the other hand, someone who just uses AI art to make their favorite anime waifus is fine, but I would never engage in them in conversation about their "art," because their art is shallow and uninteresting, even if it technically checks all the boxes. Also for what it's worth, I know what you're referring to with respect to the additional pixiv from trad artists, but the shadow work on a lot of these are actually fine. They're consistent and they make sense given their placed light source. In fact, if you're talking about niche subjects, a lot of these links have more individual styles than the AI art you showed me. This is what I mean by the fact that I think these artists, even if they are not as technically proficient with their line art, their textures and layered coloring, have more individuality and charm. On the other hand, this AI work actually reminds me of a lot of Riot art, which is technically near perfect, but lacks soul. Finally, the cool thing is based on the artist's other works you can track improvements in their line art, coloring, expressions, etc. These tell a story about how the artists recognize their weaknesses and work to improve their craft. AIs do it too, I won't deny them that, but again, to ask how they improved or what they improved as an aesthetic or artistic question would be a fruitless exercise. It's a technological marvel, and I leave that ogling to the technophiles. In addition, I think comparing AI art to amateur traditional artists is a bit of a poor comparison. But granted, let's just say for instance that AI art is superior to 40% of the art that people make, but you yourself have only given us 27/900 images, that's something like 3% of the sample size that's usable. And it's not like I haven't played around with AI generation either, it takes quite a bit of work to create an actual finished product that seems passable. I understand your point that things are going to be better in the future, and I'm pretty long this stuff myself, my point is mostly just presently the artwork is not there yet. |
YudinaFeb 26, 2023 8:14 PM
Yudina said: To be blunt if you believe AI generated art with a full techstack is not superior to the 40% I'm not sure what you're thinking lmaoCneq said: I don't actually agree that a lot of AI generated art is necessarily superior. A lot of what you showed me for instance, specifically the ones with guns or weapons, actually remind me of a lot of early 2000s art that I would find on pixiv / danbooru, and I actually find that the way they're drawn is really derivative and uninspired in comparison. Now, granted, there's an argument to be made that technically speaking a lot of AI art shows a lot of promise and yeah a ton of artists can't stack up against this stuff, but part of me also just thinks what makes art great is in the purpose that we can derive out of why someone made a particular stroke or color selection.Also as I've said the current issues with hands and anything more complex [like guns] are a temporary setback that all AI generative works cannot cope with and besides inpainting these things over and over again to achieve a desired result it will take time [which I suspect will be another 1-3 years]. When it comes to lighting and shadows before orangemix and other new "mixes" lighting was a serious issue but even with these mixes being relatively new [and more being made all the time] this issue is almost gone even with basic prompting and zero inpainting/outpainting or control net tweaks. Such as in this case: https://www.pixiv.net/en/artworks/105726514 I could find hundreds of examples of bad lighting/shadows from amateur traditional artists and yet most people using orangemix or something similar the lighting far and wide is superior. Will it be superior to top tier artists in this state? No. However what number of top tier artists exist in the current population? and how many of those people are available to cater to specific niches? characters and ideas? Hardly any. Thus in the case AI generated art undoubtedly still wins since it can create art for more niche characters/themes [which are fundamentally lacking] and likewise is superior to most art that 40% of people make. Purely on the first page alone for this character most traditional art is simply lacking: https://www.pixiv.net/en/artworks/105428168 https://www.pixiv.net/en/artworks/105148658 https://www.pixiv.net/en/artworks/105191762 https://www.pixiv.net/en/artworks/105131035 https://www.pixiv.net/en/artworks/105132253 I would personally rather view AI generated art compared to these and I would struggle to see why anyone would think otherwise lol So to take the additional pictures you posted, the reason why the AI decided to draw light above the character's nose when it makes no sense from a shadow perspective is fascinating as a technological question, but it doesn't interest me from an artist or aesthetic perspective. There is, at least not in the moment, no aesthetic question that I can ask the AI that will return a comprehensible answer with respect to art, technique, and method. That would be up to the artist who is creating the AI art and that could be an interesting discussion. And again, I think that's going to be the interesting future, what do artists do with AI and appropriate it for their needs and vision. On the other hand, someone who just uses AI art to make their favorite anime waifus is fine, but I would never engage in them in conversation about their "art," because their art is shallow and uninteresting, even if it technically checks all the boxes. In addition, I think comparing AI art to amateur traditional artists is a bit of a poor comparison. But granted, let's just say for instance that AI art is superior to 40% of the art that people make, but you yourself have only given us 27/900 images, that's something like 3% of the sample size that's usable. And it's not like I haven't played around with AI generation either, it takes quite a bit of work to create an actual finished product that seems passable. I understand your point that things are going to be better in the future, and I'm pretty long this stuff myself, my point is mostly just presently the artwork is not there yet. Also the very act of multiple generations [i.e one out of 60] is not because all of them are bad but due to personal preference of what the person generating wants. Many [if not all] can be fixed with very minor inpainting and some may appeal to different people [this is once again a human element in the creation process]. This was not the case in early 2020 AI generated art and obviously not the case with early GANs. Generation has already reached a point where most art is passable with minor fixes [and many would surpass a generation that turned out correctly the first time]. For example in the case of the Roxy image that is 100% from prompts and a single lora concept [being roxy herself]. Zero inpainting/outpainting was done nor any use of control net or other extensions and it was generated amongst 12 total generations [in which most of those are good, with only 4 requiring minor inpainting for the hands for example]. If you were to dedicate the same amount of time a traditional artist would take to make such an image and instead used that time to perfect the prompts used for that image, focused on inpainting/outpainting, used more LORA concepts for further stylization/improvements and even used control net to perfect the pose and other aspects you would most likely spend 2 hours max where as a traditional artist would a day if not much longer. Also as I've said before it will take years to "improve" yet just in the last year, or heck six months alone that improvement has already happened. There is a complete lack of art for niche game series/themes such as mushoku tensei, the trails series, muv luv, umineko and other more obscure franchises and there simply isn't any art, period. The fact AI generated art can help cater to these franchises and the fact that it's without a doubt superior to the 40% of amateur artists that exists is a net positive to both these franchises and to art enjoyers. What currently exists is certainly good enough [and far superior to what was possible in 2020] and can already fulfil certain use cases. Also: Yudina said: On the other hand, someone who just uses AI art to make their favorite anime waifus is fine, but I would never engage in them in conversation about their "art," because their art is shallow and uninteresting, even if it technically checks all the boxes. You do realize this is literally the case with 40%+ of most artists, correct? To assume this belief you would also have to classify these people as "not being artists" because their work is objectively trash, hard on the eyes and not interesting in the slightest. You seem to hold opinions that fundamentally contradict each other and feels like they all originate from the idea that AI generated art is somehow using some "cheat code" that places it below even the most horrendous traditional art [despite advanced level AI art taking far more technical skill, time and conscious vision to create]. Also, good lord: " but the shadow work on a lot of these are actually fine. They're consistent and they make sense given their placed light source." My brother in christ, are you seriously going to sit here and say this is superior? https://www.pixiv.net/en/artworks/105428168 Are you completely oblivious to the fact ANY style can be trained? The reason most of the generations here are quote on quote " soulless riot art" is because I personally enjoy that aesthetic and not because that is all AI art is capable of. If I were to scrape deviant art and pixiv for the worst looking amateur art I could find and train a dreambooth and/or lora model with that I guarantee you it would come out nearly idenitcal. I mean just look at this whacky model someone made for similar meme worthy purposes which was trained on the worst low quality art they could find: https://civitai.com/models/12910/shitty-oekaki-jaggy-lines-style Likewise fixing textures and colouring is one of the things inpainting does well and not a single generation here has used inpainting [yet almost any issues can be solved with enough time in the same way a traditional artist could]. |
CneqFeb 26, 2023 8:38 PM
|
| AI generated art gets a like. Artists who draw things on their own gets a follow. |
 |
| Man I really hate this new quote format on MAL. To be blunt if you believe AI generated art with a full techstack is not superior to the 40% I'm not sure what you're thinking lmao That's not really what my point was.You do realize this is literally the case with 40%+ of most artists, correct? To assume this belief you would also have to classify these people as "not being artists" because their work is objectively trash, hard on the eyes and not interesting in the slightest. I'd actually raise this to like 80-90% of people, but yes, you are correct, that is actually what I am saying. I'd say the same with 80-90% of novelists, and poets, and writers in general, etc. etc. I'm not sure why you think this is a controversial take. You making AI art to reflect the kind of art you want is great, and that sounds awesome. But if I were to do the same, I don't think I'd expect anyone to engage me intellectually on my process because I don't think I made anything interesting. Again, a technological marvel though it may be, I leave that to people who actually care about the tech. It's not an artistic conversation. You seem to hold opinions that fundamentally contradict each other and feels like they all originate from the idea that AI generated art is somehow using some "cheat code" that places it below even the most horrendous traditional art [despite advanced level AI art taking far more technical skill, time and conscious vision to create]. That actually isn't what I'm saying at all. I don't really think AI art is a cheat code at all, I think if anything it's just another helpful resource that's going to really help artists produce the vision that they want. It's fine to accuse other people of that, but please don't regurgitate talking points if I''ve never mentioned anything of the sort.Also, technical skill, time, and "conscious vision" are not only vacuous concepts that make almost no concrete sense (even technical skill can be highly relative depending on time period and fads), they are ultimately irrelevant. I don't judge something based on how hard it was to create. If a writer spent 10 years writing a shitty novel, then congratulations, they wrote a shitty novel. I'm not sure what 10 years of conscious vision towards a garbage product is supposed to do to convince me otherwise. This is why I don't actually care about AI. I'm not threatened by it, I welcome it. If it makes things easier for people to create the art they want, then there's no reason why we should stop this from happening. My brother in christ, are you seriously going to sit here and say this is superior? https://www.pixiv.net/en/artworks/105428168 My original point was with respect to lighting, and you simply chose a bunch of art you didn't like and posted it. And I'm telling you the shadow work here is actually fine on a technical level. It makes sense given the light source and way the coat is thrown over the character's shoulders. Is it elementary and not super professional? Sure, probably, but to compare it something drawn in a completely different style seems inane and pointless. More importantly, it just makes you seem really miffed that people don't share your objective enjoyment in the arts or that you have a personal vendetta against traditional artists, most of whom are just hobbyists or young or learning btw, and not actually excited about the prospect of AI art in general.Yes, you can say which one you like more, but this one doesn't really have any problems with shadows and lighting. Whether you actually like it or not is irrelevant. Are you completely oblivious to the fact ANY style can be trained? The reason most of the generations here are quote on quote " soulless riot art" is because I personally enjoy that aesthetic and not because that is all AI art is capable of. Yeah, I'm sure you can train any style, and if you like soulless Riot art then that's your prerogative, but that's precisely my point. If you wanted to compare AI art without being condescending to art that you simply don't like, you should instead choose examples of art pieces that are going for the same style as the ones you presented. In addition, in the same way that you're scoffing at the artists for screwing up this or that, you shouldn't be allowed to deflect from the fact that AI art is still deficient in numerous respects. Will it get better? Yeah, very likely, but call things for what they are.I mean just look at this whacky model someone made for similar meme worthy purposes which was trained on the worst low quality art they could find: I mean again, this raises interesting technical questions and questions of judgment on part of the person doing the training and not the AI itself. Because what the person doing the training is doing is making hard aesthetic and value judgments of what they want to create and I think what they tell us will reveal their preferences for what is and isn't "good" art.https://civitai.com/models/12910/shitty-oekaki-jaggy-lines-style And so, my original point remains the same. AI is interesting to me as a tool that artists can use. I think from that perspective, it has so many awesome applications, but as a tool in and of itself, it does not interest me artistically or aesthetically. It cannot answer deep and probing aesthetic questions of method, technique, and purpose. I'd honestly look forward to a day where that can actually happen, but in the meantime, I'll settle for the artist commanding the tool to tell me on its behalf. |
YudinaFeb 26, 2023 9:15 PM
| A response from a previous thread I responded to regarding it: TheAngryNerd said: I disagree with the idea of it being "art". An AI isn't a living thing, it can't create things with meaning because it has no intention, will, or a creative mind of its own. It creates things without heart, reason or thinking. I believe that by grouping AI and human art together, you're essentially objectifying both, because then it isn't about meaning, but rather about how "good" something looks over the other. This is already being done with real human-made art, let's not push it anymore. I guess it's only real ethical use would be to use it for references, since it's something real artists do all the time. Have an AI make something for you, and you use what it produces as a reference for your work. That way you're still technically making the art yourself. When I see art and realize it's AI generated it loses my interest and becomes unimpressive to me. I'd rather look at something that has actual hard work put into it through someone's years of learning and mastery. Come back to me when the AI has the ability to make art on its own without internet access or a database of other people's work to use. As for the AI illustrations that were generated for you, they're pretty, but I still don't think they're art. Here are two videos I agree with (the first one kinda devalues illustrators a bit but is nonetheless great in my opinion). |
 |
Even with stable diffusion in it's relative infantile state it literally has the ability to utilize all art/styles/concepts/ideas and to create/innovate on concepts that LITERALLY haven't been combined before lmao man's tryna tell me that's innovating. brb, gonna tip a tin of heinz tomato soup into a tin of heinz cream of mushroom and tell the world i've invented a new soup, lemme see how well that goes over. AI cannot create, it cannot innovate, it cannot make anything beyond its input. Combining several inputs to give the illusion of novelty still doesn't make the thing it spits out an innovation. It makes it a combination of existing patterns. This is also not the crux of my argument, and I find it very funny that OP chose to completely ignore my actual critique of AI image generators in their response to my argument. Nothing like a strawman to help you win an internet fight amirite. |
syrcus said: Are you blind as well as ignorant? Go look at my reply Yudina when it comes to the legality of AI generation and the fact that it isn't "theft" like your braindead ass seems to believe. Even with stable diffusion in it's relative infantile state it literally has the ability to utilize all art/styles/concepts/ideas and to create/innovate on concepts that LITERALLY haven't been combined before lmao man's tryna tell me that's innovating. brb, gonna tip a tin of heinz tomato soup into a tin of heinz cream of mushroom and tell the world i've invented a new soup, lemme see how well that goes over. AI cannot create, it cannot innovate, it cannot make anything beyond its input. Combining several inputs to give the illusion of novelty still doesn't make the thing it spits out an innovation. It makes it a combination of existing patterns. This is also not the crux of my argument, and I find it very funny that OP chose to completely ignore my actual critique of AI image generators in their response to my argument. Nothing like a strawman to help you win an internet fight amirite. It's insane how people can find time to be opinionated on something as complex AI art and yet be completely oblivious to how stable diffusion and training actually works. I swear there has to be some serious overlap with artists being the most pissed off and hostile to this disruptive technology and yet by and parcel of being so invested in art they usually do not have a strong technical background when it comes to topics such as computer science and artificial intelligence. With that it would make perfect sense as to why so many people are so incredibly off the mark when it comes to the reality of what is actually being done when it comes to stable diffusion and training [which is not theft at all]. Also are you seriously here trying to compare canned soup to one of the most complex models ever conceived with 890 million parameters which was trained on 160 million images purely for the base model? [not even counting the fact this continually grows the more and more mixes come out]. " AI cannot create, it cannot innovate, it cannot make anything beyond its input." What is creation? What is innovation? What is input? You are making sweeping generalizations based on words you seem to have no understanding of. Do you have any idea the actual process of a human artist? A human artist can't do jack shit without input. That input, much like a diffusion model, is a result of learning and for art that means seeing art and all concepts associated with it [which usually are a result of life experience] and then making associations between those concepts and what makes good art to give form [generate/create] to what was once physically formless. The process besides life experience is exactly the same especially when it comes to understanding and breaking down concepts to their fundamental components and likewise learning and making associations between these components. Both human and AI generators learn concepts, style, best practises and use these to CREATE and much like actual innovation in art the fusion of new styles based on concepts/blends of old styles/ideas is the very definition of innovation and this is the case regardless of the TOOL a HUMAN uses to innovate. I have to ask once again, do you even have the ability to read? Because you still seem oblivious to what control net and inpainting/outpainting actually are and the fact they literally give nearly the same amount of control a digital artist using a tablet in photoshop would have. You seem ignorantly stuck in the past of 2020-2021 era generation heck even 2022 when in reality the process of creating AI art has fundamentally and drastically changed since then. @TheAngryNerd " When I see art and realize it's AI generated it loses my interest and becomes unimpressive to me." A great example of my point. People such as yourself are completely ignorant on the process involved in actually creating AI art due to an outdated and ignorant idea of what "AI art" actually is and the actual process involved in making it [mostly from the idea of what it was from the 2017-2020 era]. I highly implore you to download stable diffusion and attempt to generate something that corresponds exactly to your vision and see how that turns out for you. You are ignorant to the actual workflow and skill required to make competent AI art and considering control net literally released in the last month in full you are constantly going to be ignorant due to the rate of innovation coming to these tools. I still find it hilarious people are somehow completely ignorant on the human process in AI art workflows [which includes literal drawing and photoshop integration in full] on top of all the traditional steps involved in stable diffusion [such as prompt engineering with weights, textual inversion, lora concepts and everything in between]. I wonder how many years it will take until this ignorance ceases? Perhaps when photoshop finally rebrands itself to fully take advantage of diffusion models and other similar tools in the next decade to make it mainstream to some extent. |
CneqFeb 27, 2023 1:55 AM
|
deg said: Not to mention for voice acting as well. I've trained 15 voices with eleven labs in the last month and almost all of them are a near 1to1 with the original voice actor.ye its inevitable funny that few years ago people say that Art will never be automated but here we are and its beginning especially with DeepFakes in both video and audio media just like reading and writing are once rare so highly valued then Art with AI will be devalued more because of how widespread and mainstream it will become Will this replace voice actors? Of course not. Yet it will make amateur voice actors completely redundant and likewise instead of hiring a voice actor by the hour to voice every line you can instead simply get 1 hour of voice lines in a strict and well varied data sample and use that to train their voice and you will be able to use it for anything you want. Unless this is banned for ethical reasons it is a direct benefit to companies to save massive amounts of money and the act of selling/leasing your voice for a company to use [such as for a game] will be the case for a lump sum payment. And this is just with a public and consumer level API like 11labs and it being one of the first of it's kind to offer this kind of quality. Give this another five years and you will see prompt based definitions for defining how certain words are spoken, accent, emotion, tone and just about anything [which this type of prompt based system is not available on 11labs]. Many people are going to be made redundant [besides the ones of exceptional quality] and people are starting to feel fear. |
|
Cneq said: syrcus said: [cut for length]lmao man's tryna tell me that's innovating. brb, gonna tip a tin of heinz tomato soup into a tin of heinz cream of mushroom and tell the world i've invented a new soup, lemme see how well that goes over. AI cannot create, it cannot innovate, it cannot make anything beyond its input. Combining several inputs to give the illusion of novelty still doesn't make the thing it spits out an innovation. It makes it a combination of existing patterns. This is also not the crux of my argument, and I find it very funny that OP chose to completely ignore my actual critique of AI image generators in their response to my argument. Nothing like a strawman to help you win an internet fight amirite. Imagine writing all this to justify theft when you could learn to draw instead |
syrcusFeb 27, 2023 3:57 AM
syrcus said: Imaging wasting time being opinionated on topics you have no understanding of when you could get educated instead.Cneq said: [cut for length] Imagine writing all this to justify theft when you could learn to draw instead Keep it up though buddy, far easier to cope with the "waa waa AI art is theft waa waa" to make you feel better about being made redundant as every month and year goes by. Also once again showcasing your illiteracy. I can draw, hence the fact I've mentioned time and time again the importance of control net and using a tablet to gain full control over models. I can tell you have a very hard time understanding simple concepts. |
|
Cneq said: Imaging wasting time being opinionated on topics you have no understanding of when you could get educated instead. Keep it up though buddy, far easier to cope with the "waa waa AI art is theft waa waa" to make you feel better about being made redundant as every month and year goes by. Also once again showcasing your illiteracy. I can draw, hence the fact I've mentioned time and time again the importance of control net and using a tablet to gain full control over models. I can tell you have a very hard time understanding simple concepts. "I can draw" says the guy who can't even see how fucked up the anatomy in their AI-generated images is, and keeps saying it's superior to actual art. Sure, sure. Fact is, AI is a tool, and as it stands it's a tool built on unethical ground. You have still yet to offer a convincing argument as to why copyrighted works should be used without permission as training data - because there isn't one. Nobody would have an issue with AI image generators (or, for that matter, AI text generators, or AI code generators) if they weren't built on stolen work. If you gave half a fuck about AI image generation and its future, you too would be arguing the case for ethically-sound, opt-in datasets. The fact that you brush this issue aside and ignore it tells me everything I need to know about you. The fact that you jump to trying to dismiss anyone who disagrees with you is also very telling. It's also telling that AIbros turn to artists and go "ha ha ha ha ha you're just mad because you're redundant now", when in reality art is thriving, and the machine models used to create AI generated images literally cannot function without art as an input. We are not the ones who are insecure, buddy. |
syrcus said: Oh? AI is now a tool used by humans? What happened to describing it as some all-power independent entity that just poofs out art at a flick of the wrist?Cneq said: Imaging wasting time being opinionated on topics you have no understanding of when you could get educated instead. Keep it up though buddy, far easier to cope with the "waa waa AI art is theft waa waa" to make you feel better about being made redundant as every month and year goes by. Also once again showcasing your illiteracy. I can draw, hence the fact I've mentioned time and time again the importance of control net and using a tablet to gain full control over models. I can tell you have a very hard time understanding simple concepts. "I can draw" says the guy who can't even see how fucked up the anatomy in their AI-generated images is, and keeps saying it's superior to actual art. Sure, sure. Fact is, AI is a tool, and as it stands it's a tool built on unethical ground. You have still yet to offer a convincing argument as to why copyrighted works should be used without permission as training data - because there isn't one. Nobody would have an issue with AI image generators (or, for that matter, AI text generators, or AI code generators) if they weren't built on stolen work. If you gave half a fuck about AI image generation and its future, you too would be arguing the case for ethically-sound, opt-in datasets. The fact that you brush this issue aside and ignore it tells me everything I need to know about you. The fact that you jump to trying to dismiss anyone who disagrees with you is also very telling. It's also telling that AIbros turn to artists and go "ha ha ha ha ha you're just mad because you're redundant now", when in reality art is thriving, and the machine models used to create AI generated images literally cannot function without art as an input. We are not the ones who are insecure, buddy. You've already showcased how profoundly ignorant you are on these technolgies and your dumb ass probably doesn't even know why stable diffusion is called stable diffusion for fucks sakes. And yet despite this you remain opinionated despite being completely uneducated. Also brush aside legality? Are you braindead? You literally believe art is "stolen" and not learned. You are literally dismissing fundamental aspects of how these models are trained and why it is nearly identical to how humans may learn various concepts and styles of art. I assume your avoidance of these points and the hilarious redirection of saying I'M brushing it aside is once again due to your fundamental ignorance on how these models are trained (I wouldn't be surprised if you're only input on these models is from some shitty 30 minute youtube video by a biased artist trying to justify their existence). Come now buddy, explain to me the proccess of how these CONCEPTS are learned and how learning CONCEPTS equate to theft. You are the one brushing aside the issue because you fundamentally lack the knowledge to respond. " says the guy who can't even see how fucked up the anatomy in their AI-generated images is," Nice strawman dipshit, or are you just illiterate? I literally pointed out issues with features in my images multiple times and likewise clarified zero inpainting nor control net were used on this. Your dumb ass on the otherhand doesn't even know what control net nor inpainting even is yet you sit here in this thread flailling around in ignorance because you're a butthurt artist who should've went to college and got a STEM education instead. Also it's very cute how you believe more art is somehow required for AI art to continue to exist. All art both digital and not already exists and likewise can, will and has been trained on. Likewise generated art itself can be used to train. The way in which concepts are learned means 160 million is more than enough to create almost anything and by no means requires more artists. 160 million is a small model and 250 million image models exist and likewise 500million plus models also exist. The fact you seem incapable of understanding the actual quantity that 250 million images actual is (with billions of parameters) goes to show your technical ignorance. How much art have you seen in your life? 5000 pieces? 50,000? 500,000? Will you ever see 250 million, tag all of these and understand what concepts these images consist of and be able to apply those concepts at will? No. You underestimate the actual complexity and sophistication of these technologies and will be left behind within the next 10 years |
CneqFeb 27, 2023 6:18 AM
|
We are going completely off course. I'm not talking about the technical aspects, i'm talking about human behaviour and technology. We have schools dedicated to art, people follow educations for years and other people learned it themselves by dedicate their entire life to art. They develope skills to create the most beautiful art. I don't care about how a AI works or how you learn it. All I care is the human behaviour and the side effects it can create. Learning a AI how a astronaut looks like and how a horse looks like and then create a astronaut on a horse, isn't nowhere the same level as someone who makes art with their hands. People create the most insane and beautiful art. This isn't nowhere the same level as generated by AI. I will give you a good example of what is happening. Craftsmanship is disappearing, due the fact that almost everything is produced by factories, massa production. People stop doing math with their head, since we have a calculator. People stop recognize streetnames, places and environmental points of interests, because they are driving mindless on a GPS. Soon or later when the mass of the people will figure it out how AI works, they start losing their traditional traits and skills. Why would you still go to a art school, if you can create the most amazing pictures with a AI? How many glass is still handmade? How many furniture is still handcrafted? Our society's are losing skills and traits and become slowly one with machines. Many people can't live without their phone anymore, it's their most precious thing. It does almost everything for you, it's the closest companion people have. And if you have a issue, just look it up on your phone. You don't need to have skills or traits anymore to solve something yourself, just ask your goddamn phone. If you take it away from them, they will panic. They can't figure it out anything on their own anymore. Like I said, I do love technology, but I can also see what it does with humans. And i'm a person who wants to preserve tradional skill and traits, when we become slowly machines ourselves. And therefor I think that technology also needs to have regulations, including AI. There was even a news article about how students nowadays let a AI write their essay. Are we really want to go live in this kind of world? If I had the power, I would put internet offline for one day, for just one single day, I will bet for thousands of dollars that it will create a global panic attack. This is how we are now one with machines, we need them, because we are losing our own skills and traits, since a lot of things are done by machines. This is also how weak we become as a human species. That's why i'm a proponent for technology regulations. A AI should always be a secondary tool. And no, I don't believe that robots or AI will ever rule the world. But I do believe that we are losing our traditional self, and that we will lose certain traits and skills and let the machines figure it out for us. But one thing I can say certainty, it's good that we talk about these kind of subjects and it reaches all the way up to political level. And we should doing this, until we have a agreement of how we need to regulate technology. So, I would like to ask everyone to keep it civil in here. Because I can see the tension growing. |
Toonen1988Feb 27, 2023 12:47 PM
| "Most people talk about killing time while time is killing them. You can outrun everything but you'll never outrun the hands of time. Use it wisely before you expire". - Toonen1988 "Cyberpunk show us the dark side, reveiling the dangerous side effects of the drug of futurism." - Indigo Gaming |
Ingmatica said: Instead of watching videos entirely made by artists [in which their entire channels are based on art and most likely their careers] and by definition prone to extreme bias you should instead be watching technical videos that have no bias and can then use this to see the flaws in the very idea that AI art somehow "steals" and "slaps art together" when in reality the process is akin to a human learning concepts [albeit in this case at a slower more undeveloped and nuanced pace and in a very artificial way.TheAngryNerd said: A response from a previous thread I responded to regarding it: When I see art and realize it's AI generated it loses my interest and becomes unimpressive to me. I'd rather look at something that has actual hard work put into it through someone's years of learning and mastery. Come back to me when the AI has the ability to make art on its own without internet access or a database of other people's work to use. As for the AI illustrations that were generated for you, they're pretty, but I still don't think they're art. Here are two videos I agree with (the first one kinda devalues illustrators a bit but is nonetheless great in my opinion). I would also add this one for conclusion. Artists who literally have no technical background making extremely opinionated videos about a high technical topic with tons of sweeping generalizations and falsities does nothing more than to give false hope to people who rather deny reality instead of upskilling [such as integrating AI technology into their existing workflow to get ahead of the curve.] |
|
| Yuck. AI "art" can burn to the ground, for all I care. |
 |
| I think AI art is soulless, to the point whenever I see a nice looking image and then I realize it is AI generated, it stops looking good to me, don’t know if I’m being overdramatic but that’s the way it is. Also I feel sad for real artists. |
Manuel-AM said: This is purely a psychological response born from the almost fanatical and widespread delusion that AI art is somehow "theft" and "easier" than making traditional art which is nothing more than a delusional conjured by coping artists.I think AI art is soulless, to the point whenever I see a nice looking image and then I realize it is AI generated, it stops looking good to me, don’t know if I’m being overdramatic but that’s the way it is. Also I feel sad for real artists. As someone who's literally been using photoshop since 2010, used cinema 4d from 2011-2014, used sony vegas from 2012 - till present, same for after effects and likewise have also used blender for many years in almost all of these creative workflows the process and general ease of use for the end user is FAR more user friendly and easier than attempting to generate high quality AI art. I say this as someone who's literally done paid commissions for graphic design in the past and likewise has won more than 30+ physical in-person awards for video editing and the process of generating AI art is far more involved for the end user. |
|
EyeAmTheI said: The time it took you to type this you could've actually educated yourself and realized your entire opinion that AI art "does not take any effort" is complete bullshit.My stand (aka opinion) on this is that art has to require effort and desire to express something. If you generate your art with some AI than A) you are not putting any effort into it (as an artist) B) It is not expressing your desire, the AI only generates an interpretation of your imagination. Hence, the result never will be your true intentions only a rough guestimate. Not to mention, when you are generating art, you don't make art. It is more like you are browsing in an endless library while you are trying to figure out which is the correct "keyword" to find the one you are looking for. So yeah, I don't consider AI generated (or aa lot of post-modern art for that matter) stuff as an art... You can literally draw with a tablet for the entire process of AI generation and this applies to SD inpainting/out-painting as well as all modes in control net [two being native and the rest possible through the use of blender to generate similar maps]. ControlNet fully supports line drawing and likewise due to the numerous times I've explained the process of how SD models and lora concepts/textual inversion is used the complexity of the AI workflow is LITERALLY more complex than what a traditional artist has to deal with. https://github.com/lllyasviel/ControlNet Purely to perfect hands you have to use blender to create a depth map and manually integrate it which requires constant messing around with the various parameters in SD that have a tendency to screw things up in which most cases you need to literally reverse the process of noising up an image to fix these issues. https://www.reddit.com/r/StableDiffusion/comments/11d7it7/targeted_hand_fix_is_finallly_possible_at/ That is to just get a perfect hand, when it comes to perfect eyes do you think perfect eyes are simply generated by AI? Hell no. There doesn't exist a single model even running on a 4090 that can generate perfect eyes on every attempt and to fix eyes [every single time] you are required to find a perfect seed, take that seed, regenerate it until it's correct, crop out the face, take that into img2img, use an extremely high resolution on that image with inpainting only applied to the eyes with the correct sample method and prompts chosen and then do a batch of 20-30 generates [which would take a 3080ti roughly an hour to complete at this resolution] to find a good working pair of high resolution eyes that fit with the rest of the image. After this set of eyes is found, it's off to photoshop to carefully apply and crop this new set to the previous generation which may or may not work and due to the higher resolution further attention to the overall image will need to be fixed [since usually eyes are generated too bright compared to the rest of the image] and thus using something like magic bullet looks for image processing is usually a good idea. That is the process purely for eyes, that's it, just the fucking eyes. I guarantee manually drawing eyes to an image you've already created precisely to fit said eyes is FAR easier than attempting to integrate eyes into something that literally does not support it. And this same type of shit applies to any concept you may want to change, ALL of it is manually and done by the end user and ALL of it requires specialized third party tools such as photoshop, blender, magic bullet looks and of course it's almost mandatory to have gigapixel and gigapixel is yet another process in this workflow. It's honestly insane how 80% of the people opinionated on AI art literally have a perception of it that could be defined in two sentences and those two sentences are nothing but falsehoods born from misconceptions and ignorance. |
|
| Well first off I think the pictures Cneq used AI to generate look okay to me, I'm not experienced enough to pick up on the minute differences that others are talking about. Looked briefly on the link he provided and all the other images look good too. Despite that I find myself agreeing more with Toonen1988 & EyeAmTheI's point that AI Art presents a worrying future as AI generated threatens both people who made the art the AI's are using as a database and the people who are planning on making art that would be displaced by AI. Scud: I'm also not convinced that it's somehow 'unethical' - real artists learn by studying the real world and other artists right? I don't see how that's fundamentally much different from how you teach machines to do it, myself. I get the sentiment, but I think the difference is that humans have to go to school and study these things, which takes time and effort, and experience is used as inspiration for their own works. This is in contrast to AI which meshes together if not billions of pieces of data to form a composite image to fit your prompt. It's hard to compare the "effort" taken by a computer versus the effort undertaken by artists. And there are many artists who are innovators, such as DaVinci or Van Gough or maybe in the Anime context someone like Satoshi Kon, who probably were inspired by something but then created something entirely different. SynapticBlast said: My word may not mean much, but hope you keep at it. I double majored in two subjects in college that proved pretty useless in finding a job, but I still value the experiences I had and the relationships I formed with my classmates. My job now has absolutely nothing to do with what I did in school but then again that's the norm for a lot of people. Honestly kind of scary. I'm trying to learn digital art and it feels like there's not much point considering the progress AI has made in literally 2-3 years. So yeah in conclusion I know I'm needlessly butting in on the conversation, and while it's pretty cool to think of what AI can do now, and will be able to do in the future, I cannot replicate the victorious attitude Cneq has regarding this topic. Perhaps I shouldn't think about how much it will disturb the art world and think instead of how many new opportunities it will bring, sort of how CNC opened a lot of new avenues for home machinist (for those who don't know it's a machine that you enter a program into it and it basically machines most of the thing while you don't have to do much with your hands). I think this article sums up my attitude: AI Generated Art Is A Copyright, Ethical And Working Dystopia (kotaku.com) . And lastly if anyone is still reading (no one lol) I would highly recommend an Anime called "Puparia" it was made by an animator who got fed up with the industry and decided to make his own thing, completely hand drawn. It's a neat product of someone following their artistic efforst and passions. |
Ingmatica said: The majority of DALLE-2 [which was used as a cornerstone to push this bs "ai art is theft" narrative in the first place] was most likely trained entirely on royalty free stock photos. Due to the common issues DALLE-2 suffers from many people [despite the training process being closed source] have agreed this to be the case.AI training models use "open source data", right...(it's like searching google for an image and then claiming that you've created "art"), lol. OpenAI most likely partnered with stock photo companies to do this and yet despite this even then people called DALLE-2 theft despite it being even further from the case than something like stable diffusion, which, news flash, training is not theft. I've still yet to see a single person in this thread actually manage to reply to me and give reasons as to why artificially learning something [in a more odd way, yet remains true when it comes to morality] of taking input and learning concepts/ideas from this input and how they associate with one another. A traditional artist processes concept/ideas to improve and see what "works" and what "doesn't" and likewise what type of styles they enjoy and unconsciously and consciously replicating these styles into their own creations. This is literally the same exact process when it comes to stable diffusion and in reality SD doesn't even handle this since you literally need a human element for the entire workflow. @EyeAmTheI Alright, so you believe art is not at all about creative expression and giving from to the formless but instead entirely based on some abstract [and very modern] idea of "artistic effort" and that "artistic effort" entirely depends on a boolean expression of whether your art uses "true" art tools or not. If someone as irrational as you existed in the 14th century and had a firm understanding of digital art you would be saying to people of the future that none of them art putting in "artistic effort" because they aren't using your shitty 14th century ideas and tools to create and thus they are not putting in any "artistic effort" and what they create is not "art" EyeAmTheI Not to mention, it has a good chance that I have a way better understanding of how that thing you are using actually works than you are... So yeah, sometimes people would be wiser, if they wouldn't just assume the other is a complete moron.. Extremely unlikely. I'm a CS major who literally formally studies intelligent systems of all kinds and have experience with just about every current implementation of AI there is. Unless you're a postgrad computer science major that has spent the last seven years closely monitoring these space on a daily basis you are completely out of your depth when it comes to the technicality of AI generation. The fact you don't even know what control net is despite it literally changing the entire space is a very good indicator that you are not at all familiar with the current state of things [which is funny considering anyone who has worked with ANY form of AI would realize the rate of new advancement is not only on a monthly basis but literally on a weekly basis now, there is no AI winter for the last few years.] |
CneqFeb 28, 2023 2:39 PM
|
| I will go through this thread later but.. wow, I so badly wish I was good at STEM subjects and had a degree in one of them :( And, wow, AI has been exploding in the recent times... by the time you know of one thing, there is another on the horizon. I've been thinking about this, and though my understanding is very little, something crossed my mind recently. And, I was thinking if AI is, in simple terms just a function of algorithms created by human, it's ability to collate and create art when given a suggestion, also must be human in origin... ? I say this in very simplistic terms. But firstly, what are these terms, "intelligence" and "art"... what is it to "express" and "imagine" ? Whenever I think of these terms in relation to human intelligence and artificial intelligence, I wonder.. I wonder if our own intelligence also springs from some algorithms, that are not visible or tangible to us like the AI, that humans created. If it were the case, then what are terms like "art" and "imagination" ? Do they also bubble up in our consciousness from some unknown/invisible algorithmic roots in our "mind" ? If so, then what is the difference between AI and human intelligence, and thus, what is the difference between art made by either. But then, would AI feel an impulse to "express"..? To produce or create upon being given a prompt or suggestion is one thing, but to feel an impulse and instinctive drive to express "itself" ? Isn't that also how art is made..? |
silentviewFeb 28, 2023 11:42 PM
| So... @Cneq, got any drawings that you've done without using an AI? |
|
| I'm very interested in the advent of AI art, because as an artist myself I can understand some of the issues of artists, as it can reduce their livelihood, abuse in some shape or form their own works... Too much of something saturates a market, an aesthetic and a style. But at the same time I'm a huge nerd and actually want to use AI myself, deep learning or neural network or a generator model for my own homebrewing. That's technocological anarchy, that's empowering. And it could help me do something I wanted to do the hard way, being so great at art that I could influence others into producing more great works in a certain asthetic, yuri. So I guess its juridiction is the most interesting thing to await, to clear the way AI art generators can be morally acceptable, or consensually accepted in society. I'd also like to work a little bit on a art generator for my own devices, even if only a private copy which should then be relatively harmless. It's not like I value art as sacred for every piece I stumble across. |
 |
| It is what it is, if you know what I mean. |
  |
Ingmatica said: Kind of odd you seem to completely refuse to actually learn about AI generation yet seem fully fine with having your only source of information on the topic being by definition biased artists who's entire idea of what "AI" even is and the current state of it [and future of it] is completely false.*Disclaimer, AI posting here (and not me) One for dudes (cill talk) One for girls (with calling certain types of individuals Human Trash fetishes) But this young girl nailed everything, (that's it, I'm washing my hands from all this AI art nonsense, bye.) Go learn the subject, forming your entire opinion on something based entirely from opinion pieces of the side that is the most at risk is just irrational. It's like forming your entire opinion on flat earth by watching a bunch of flat earth videos by flat earth believers and not studying science or the opposing side. That type of close mindedness never amounts to anything more than deeper ignorance, ignorance even deeper than if you didn't know about the subject in the first place. |
|
Cneq said: In terms legality it's not like there is some active web scraper that grabs all art the moment it is uploaded. The majority of art [such as low quality art] is never touched since it's bad for training models and likewise most of this art is not copyrighted in the slightest and exists [and has existed] in the public domain for a very long time. In US law copyright is implicit where upon creation it’s copyrighted just not documented as such. “When your client writes song lyrics, paints a painting, chisels a sculpture, or assembles jewelry, the copyright comes into existence automatically. These are all situations where the work is a “tangible medium of expression” (which also includes computer memory).” https://attorneyatlawmagazine.com/public-articles/intellectual-property/copyright-is-automatic I wouldn’t call it theft though, that’s not best term to describe the ethical problems surrounding how images are collected and how the existence of this tech doesn’t really benefit artists as much as it harms them which isn’t inherent just a consequence of the current economics of this society. A person generating AI images with some prompts isn’t quite what Id call being an artist something like “prompt engineer” or “crafter” would be more fitting terminology but even that is only if the prompts are fully original not just copying other prompts or using image to image in which case it’s below that in such instances. That would be akin to tracing or taking photos of photos. |
traedMar 2, 2023 3:38 AM
| ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⣸⠋⠀⠀⠀⡄⠀⠀⡔⠀⢀⠀⢸⠀⠀⠀⡘⡰⠁⠘⡀⠀⠀⢠⠀⠀⠀⢸⠀⠀⢸⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠁⠀⣀⠀⠀⡇⠀⡜⠈⠁⠀⢸⡈⢇⠀⠀⢣⠑⠢⢄⣇⠀⠀⠸⠀⠀⠀⢸⠀⠀⢸⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⢰⡟⡀⠀⡇⡜⠀⠀⠀⠀⠘⡇⠈⢆⢰⠁⠀⠀⠀⠘⣆⠀⠀⠀⠀⠀⠸⠀⠀⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠤⢄⠀⠀⠀⠀⠀⠀⠀⠀⡼⠀⣧⠀⢿⢠⣤⣤⣬⣥⠀⠁⠀⠀⠛⢀⡒⠀⠀⠀⠘⡆⡆⠀⠀⠀⡇⠀⠀⠇⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⢵⡀⠀⠀⠀⠀⠀⡰⠀⢠⠃⠱⣼⡀⣀⡀⠀⠀⠀⠀⠀⠀⠀⠈⠛⠳⠶⠶⠆⡸⢀⡀⣀⢰⠀⠀⢸ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⣀⣀⣀⠄⠀⠉⠁⠀⠀⢠⠃⢀⠎⠀⠀⣼⠋⠉⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠴⠢⢄⡔⣕⡍⠣⣱⢸⠀⠀⢷⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⡰⠃⢀⠎⠀⠀⡜⡨⢢⡀⠀⠀⠀⠐⣄⠀⠀⣠⠀⠀⠀⠐⢛⠽⠗⠁⠀⠁⠊⠀⡜⠸⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀ ⢀⠔⣁⡴⠃⠀⡠⡪⠊⣠⣾⣟⣷⡦⠤⣀⡈⠁⠉⢀⣀⡠⢔⠊⠁⠀⠀⠀⠀⢀⡤⡗⢀⠇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⢀⣠⠴⢑⡨⠊⡀⠤⠚⢉⣴⣾⣿⡿⣾⣿⡇⠀⠹⣻⠛⠉⠉⢀⠠⠺⠀⠀⡀⢄⣴⣾⣧⣞⠀⡜⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ ⠐⠒⣉⠠⠄⡂⠅⠊⠁⠀⠀⣴⣿⣿⣿⣿⣻⣿⣿⡇⠀⠀⢠⣷⣮⡍⡠⠔⢉⡇⡠⠋⠁⠀⣿⣿⣿⣿⣄⠀⠀⠀⠀ |
| From what I've seen, including your pixiv account, AI art always looks like shit. You can immediately tell when art is AI generated, because its completely bland and lifeless. Its honestly insane how unremarkable and generic it is. The second I look away, i can barely remember what it looks like. Art conveys feelings and the individuality of the artist which is something that AI, so far at least, cannot do. Even if it takes time to learn how to utilize AI for art, it's nothing compared to the years of practice that real artists go through, which always shows through their work. The majority of AI art is boring and tasteless, and the sense of accomplishment and satisfaction from creating a piece of art with your own hands and mind just isn't there. Personally I believe that AI is much more suited as a tool for artists to provide inspiration, rather than something that creates art itself. |
lain99999 said: Considering I've literally gained 1005 followers on pixiv and 6000+ bookmarks and 4000+ likes within the two weeks I posted and literally touched stable diffusion for the first time the idea that my art looks "shit" is completely subjective and most likely entirely defined by your own fragility of being an artist and feeling threatened by technological advancement. From what I've seen, including your pixiv account, AI art always looks like shit. You can immediately tell when art is AI generated, because its completely bland and lifeless. Its honestly insane how unremarkable and generic it is. The second I look away, i can barely remember what it looks like. Art conveys feelings and the individuality of the artist which is something that AI, so far at least, cannot do. Even if it takes time to learn how to utilize AI for art, it's nothing compared to the years of practice that real artists go through, which always shows through their work. The majority of AI art is boring and tasteless, and the sense of accomplishment and satisfaction from creating a piece of art with your own hands and mind just isn't there. Personally I believe that AI is much more suited as a tool for artists to provide inspiration, rather than something that creates art itself. This is especially true when you consider pixiv literally hides AI art and the stigma associated with it. What happens when that stigma is gone and AI art can be seen without a label of being AI generated? The very delusion that AI art looks any different than "real" art is going to be faded because it is nothing more than a psychological response by a vocal minority that feels threatened. Also, my art has been entirely AI generated and since then many lora concepts have been released that actually fix most of the common issues with AI art such as hands, guns, lighting and other aspects AI often struggle with you will find in 2-3 years time AI art will be free of the last bastion of cope people hold onto. Thankfully in my case however I can do your job and same can be said many others who can use AI within a fraction of the time and yet still have the technical skills to make a higher salary than you will in your entire life. Do you even realize how much it costs to commission an artist to design UI elements? Logos? Game assets? Background art? 2D character models? I can literally do all of that and same can be said for many others and that is the reality of it. I'd suggest confronting reality and realize in 2-3 years time 50% if not more of a professional artists workload will be made redundant. |
|
More topics from this board
» Missing Person - [Higurashi: When They Cry/VHS] - ProseRenaShionFTW - Yesterday |
6 |
by RenaShionFTW
»»
10 hours ago |
|
» 【 ART THREAD 】Let's share our art! ❤︎ ( 1 2 3 4 5 )mewmewforever - Aug 30, 2024 |
221 |
by Retro8bit
»»
Yesterday, 8:13 PM |
|
» Share Your YouTube Channel/Videos! ( 1 2 3 4 5 ... Last Page )nin-tendo - Dec 16, 2022 |
606 |
by Grim34
»»
Yesterday, 6:16 PM |
|
» colourful modern list (adapted from Clarity) ( 1 2 )sunnysummerday - Oct 14, 2024 |
82 |
by Retro8bit
»»
Oct 26, 11:34 PM |
|
» Four Ages! ( 1 2 )Robert_SS_Gordon - Feb 21 |
54 |
by Retro8bit
»»
Oct 24, 9:03 PM |