7
What are the limits of AI art?ev_station
Okay so please read this before you comment!
I dont want to start an entire fight in the comment section on like the ethics and studf about AI art!!
Im only curious about something that i think is really interesting which is like what are the like technical or artistic limitations of ai art? Like im not talking about "it doesnt have the human touch" im purely speaking in terms of like technical capability like "it cant draw hands" or stuff like "it cant replicate exact human proportions" like teechnical or like artistic stuff
The reason i bring it up is because i thought maybe there exist some art form like impressionism or pointism which are like SUPER difficult for any ai to recreate and then i was like what is it that makes a human able to do that???
Again please dont fight about ethics and stealing and all that, i just want to talk about what the technical capabilities limits of an ai art is??
Oh and like do you like think there existss artstyles that ai cant do because its beyond the capabilities of it?
I dont want to start an entire fight in the comment section on like the ethics and studf about AI art!!
Im only curious about something that i think is really interesting which is like what are the like technical or artistic limitations of ai art? Like im not talking about "it doesnt have the human touch" im purely speaking in terms of like technical capability like "it cant draw hands" or stuff like "it cant replicate exact human proportions" like teechnical or like artistic stuff
The reason i bring it up is because i thought maybe there exist some art form like impressionism or pointism which are like SUPER difficult for any ai to recreate and then i was like what is it that makes a human able to do that???
Again please dont fight about ethics and stealing and all that, i just want to talk about what the technical capabilities limits of an ai art is??
Oh and like do you like think there existss artstyles that ai cant do because its beyond the capabilities of it?
Create an account or sign in to comment.
71
they cant draw a oc made by a 3 year old in a maid suit on a jog to the local cafe on mars
Doing some reading on this subject and came across the following:
AI Company Resurrecting Classic Hollywood Stars to Read Audiobooks
While this involves other than visual art, it does seem relevant to the root question. [Among other possibilities, one might be able to employ a simulacrum of the original voice actor for such iconic characters as Bugs Bunny.]
AI Company Resurrecting Classic Hollywood Stars to Read Audiobooks
While this involves other than visual art, it does seem relevant to the root question. [Among other possibilities, one might be able to employ a simulacrum of the original voice actor for such iconic characters as Bugs Bunny.]
AI can't make everything like a human would because it's a mix of what humans make. For example, an AI image generator has strange text because it's just mixed text humans have made. That's also the same with AI art. You can't imitate something perfectly until you understand how it works. An AI can't understand everything that every single image has because it doesn't have all information in the world.
To the best of my knowledge, AI has some flaws in the very method it uses to generate images (I believe that it uses noise maps? (Please correct me if this is wrong)) which would need to be sorted out before I'd get truly worried that AI was going to fully replace artists. I'm specifically referring to the 5 art fundamentals of
1. Composition
While AI can without a doubt do very simple compositions, its still merely guessing, meaning that it technically doesn't know when to use for example triangles, circles or squares, meaning that while its not always the case, AI in my experience, seems very prone to just ignore basic principles of composition, meaning that getting it to generate anything beyond a very simple composition, usually yields some pretty chaotic results
2. Value / Light
Since the AI generates images based off of guesses (guesses that, tbh can be incredible!), it equally seems prone to make a lot of value and lighting mistakes. Basically: If you take an AI generated image, and turn it black/white – I've noticed that it sometimes ends up becoming very messy with no inherently clear shapes (obviously there are exceptions, this is just through testing image generation thorugh ChatGPT and GenCraft, so its difficult to say "Its all AI!"). Theres equally the problem that – again, it's some incredibly amazing guesses it can make! – it doesn't know where to put the lightsource, so it usually has multiple conflicting lightsources that ruin the coherency of the image (though the chaotic value system actually kinda saves it here, because if the values are already a mess, it gets really difficult to tell that there might be something wrong with the light).
3. Color
Same issues as with value/lighting
4. Form and Perspective
In my own testing, this is deftly where AI image generation struggles the most. Its amazing at doing simple and plain perspective, and its even greater at portraying specific mathematical shapes (at least ChatGPT is, GenCraft couldn't figure it out). But ask it to do a simple task of drawing 3 boxes in a 2 point perspective (something a beginner artist can do with ease), and the AI kinda just poops itself. I think this is due to the way it generates images using noise maps, and I don't believe that AI will ever be able to do more complex perspective than its current state BEFORE the people behind the AI, manage to find either a work-around, or a new approach that does not use noisemaps.
Edit:
A pretty good example of what I mean is this image (human drawn). From my testing, AI has the least trouble doing long lens imagery (assuming we ignore that the background always seem completely out of perspective), while asking it to do specific distorted perspective is where the AI just kinda gives up and can't ever seem to do a proper job with consistent proportions to the perspective.
5. Technical ability with tools
This one kinda goes back to my main point: I think the limits of AI image generation is in the use of noise maps. There may be a work-around in these if a human artist used AI image generation as a tool, not really acting as an artist anymore, but more like a director, who goes in and builds the framework that the AI can't create by itself, and then sort of like planting those garden-sticks that stuff like tomato-plants grow around – letting the AI "grow" around that framework. That might be a potential work-around, but will still require a human artist (or director), to correct the mistakes that stem from the use of noise maps (see the 4 first art fundamentals)
1. Composition
While AI can without a doubt do very simple compositions, its still merely guessing, meaning that it technically doesn't know when to use for example triangles, circles or squares, meaning that while its not always the case, AI in my experience, seems very prone to just ignore basic principles of composition, meaning that getting it to generate anything beyond a very simple composition, usually yields some pretty chaotic results
2. Value / Light
Since the AI generates images based off of guesses (guesses that, tbh can be incredible!), it equally seems prone to make a lot of value and lighting mistakes. Basically: If you take an AI generated image, and turn it black/white – I've noticed that it sometimes ends up becoming very messy with no inherently clear shapes (obviously there are exceptions, this is just through testing image generation thorugh ChatGPT and GenCraft, so its difficult to say "Its all AI!"). Theres equally the problem that – again, it's some incredibly amazing guesses it can make! – it doesn't know where to put the lightsource, so it usually has multiple conflicting lightsources that ruin the coherency of the image (though the chaotic value system actually kinda saves it here, because if the values are already a mess, it gets really difficult to tell that there might be something wrong with the light).
3. Color
Same issues as with value/lighting
4. Form and Perspective
In my own testing, this is deftly where AI image generation struggles the most. Its amazing at doing simple and plain perspective, and its even greater at portraying specific mathematical shapes (at least ChatGPT is, GenCraft couldn't figure it out). But ask it to do a simple task of drawing 3 boxes in a 2 point perspective (something a beginner artist can do with ease), and the AI kinda just poops itself. I think this is due to the way it generates images using noise maps, and I don't believe that AI will ever be able to do more complex perspective than its current state BEFORE the people behind the AI, manage to find either a work-around, or a new approach that does not use noisemaps.
Edit:
A pretty good example of what I mean is this image (human drawn). From my testing, AI has the least trouble doing long lens imagery (assuming we ignore that the background always seem completely out of perspective), while asking it to do specific distorted perspective is where the AI just kinda gives up and can't ever seem to do a proper job with consistent proportions to the perspective.
5. Technical ability with tools
This one kinda goes back to my main point: I think the limits of AI image generation is in the use of noise maps. There may be a work-around in these if a human artist used AI image generation as a tool, not really acting as an artist anymore, but more like a director, who goes in and builds the framework that the AI can't create by itself, and then sort of like planting those garden-sticks that stuff like tomato-plants grow around – letting the AI "grow" around that framework. That might be a potential work-around, but will still require a human artist (or director), to correct the mistakes that stem from the use of noise maps (see the 4 first art fundamentals)
I have to disagree with you because it depends on what data you train it on. I agree that it needs a sort of framework, but there are artists (mainly on Twitter, and for some reason like, mainly Chinese??), which have trained AI capable of generating images like these
Wauw! Ngl, I am genuinly impressed at how well these look o:
I can agree with you that my point about distorted perseptive seems to have been off, since it seems that the illustrations showcased, would deftly pass for some of the mistakes merely being products of a distorted perspective. However, my main point still stands that the method of generating the image seems flawed due to its lack of fundamentals.
In the fish-eye lense image, the background (which is a key role of the image itself), is sort of like this pretty well known image (I think?):
I think it was mentioned in a Vsauce video, where Stevens explains that, at a first glance all of these things look recognizeable, but the moment we start to think and look closer, we realize that nothing in the image, actually is recognizeable – yet at the same time, it's not completely unrecognizable. I feel the same effect with the fish-eye lense example provided – at a first glance there is no problem, it looks like a distorted inside of a house, but the moment I start to think about whats actually going on in the image, it starts moving into uncomfortable territory.
As for the second image – it suffers from my main point again: If the AI did not use noise maps to generate the image, then it would've been able to tell the sword from the eye – but with the use of noise maps, it can't tell the two apart because its essentially like painting on 1 layer, so the eye starts bleeding into the sword. The sword itself also falls under my point about value: There are no comprehensible shapes to make out in the base of the sword, which kinda again, goes back to the image I linked to: Difficulty with telling whats in / going on, in an image, usually leads humans to feel deeply uncomfortable.
This has more to do with humans than the AI itself, but it seems that when people notice something that is even slightly off in an image, every time we come back to set image, we can't help but constantly look at that specific error. I myself have had images in my references folder, where I all of a sudden noticed a very slight detail being off, only to realize that it is an AI generated image I've been looking at.
I know this is moving outside what Greg pointed to being the purpose of this forum discussion – but it seems sort of like AI image generation is more about fooling humans, than making something humans can appreciate. I can already hear all the AI-bros shouting "cope!" in the back of my mind of stating this haha
I can agree with you that my point about distorted perseptive seems to have been off, since it seems that the illustrations showcased, would deftly pass for some of the mistakes merely being products of a distorted perspective. However, my main point still stands that the method of generating the image seems flawed due to its lack of fundamentals.
In the fish-eye lense image, the background (which is a key role of the image itself), is sort of like this pretty well known image (I think?):
I think it was mentioned in a Vsauce video, where Stevens explains that, at a first glance all of these things look recognizeable, but the moment we start to think and look closer, we realize that nothing in the image, actually is recognizeable – yet at the same time, it's not completely unrecognizable. I feel the same effect with the fish-eye lense example provided – at a first glance there is no problem, it looks like a distorted inside of a house, but the moment I start to think about whats actually going on in the image, it starts moving into uncomfortable territory.
As for the second image – it suffers from my main point again: If the AI did not use noise maps to generate the image, then it would've been able to tell the sword from the eye – but with the use of noise maps, it can't tell the two apart because its essentially like painting on 1 layer, so the eye starts bleeding into the sword. The sword itself also falls under my point about value: There are no comprehensible shapes to make out in the base of the sword, which kinda again, goes back to the image I linked to: Difficulty with telling whats in / going on, in an image, usually leads humans to feel deeply uncomfortable.
This has more to do with humans than the AI itself, but it seems that when people notice something that is even slightly off in an image, every time we come back to set image, we can't help but constantly look at that specific error. I myself have had images in my references folder, where I all of a sudden noticed a very slight detail being off, only to realize that it is an AI generated image I've been looking at.
I know this is moving outside what Greg pointed to being the purpose of this forum discussion – but it seems sort of like AI image generation is more about fooling humans, than making something humans can appreciate. I can already hear all the AI-bros shouting "cope!" in the back of my mind of stating this haha
Good points and well expressed.
I suspect that to which you allude in point #5 may represent one branch of future fine art, with expert-systems/'AI' being used by human artists in somewhat the same role as the old masters used apprentices. [Doing the leaves on the tress in the BG or the bricks in a wall seems the sort of thing many artists would happily outsource — provided they retained aneditorial veto on the attempts.]
I suspect that to which you allude in point #5 may represent one branch of future fine art, with expert-systems/'AI' being used by human artists in somewhat the same role as the old masters used apprentices. [Doing the leaves on the tress in the BG or the bricks in a wall seems the sort of thing many artists would happily outsource — provided they retained aneditorial veto on the attempts.]
So ai not know art fundamentals that is good to know and sound very much like big technical limit of ai very interesting
It can really nail down rendering for ai art but when it comes to literally anything else it tends to mess up aka AI Hallucinations
You could see it change art styles in every part of the subject and it can mess up on even beginner stuff.
Pixel art or Voxel art has AI having insanely hard time creating, but there is lots of controversy with AI art and its best to know how to catch it when it does pop up
Some helpful tips for catching ai art are
Reflections
Joints
Eye's
Line Work
Rendering
Bleeding Pixels
You could see it change art styles in every part of the subject and it can mess up on even beginner stuff.
Pixel art or Voxel art has AI having insanely hard time creating, but there is lots of controversy with AI art and its best to know how to catch it when it does pop up
Some helpful tips for catching ai art are
Reflections
Joints
Eye's
Line Work
Rendering
Bleeding Pixels
Okay thank you so ai is like really bad at pixelart maybe is because it doesnt know how to make clear borders??
its definitely better than an average skilled artist
What does it have to do with technical limit??
If a cheap (i.e. automated) alternative esists that will give a product qt least equal in technical merit to that of "an average skilled artist", one would hire only artists with technical skills sufficiently above average that the increased cost was justified by increased quality.
(There is, however, more to [non-commercial] art than technical skill; a human artist of superior vision/passion/etc might still find a ready audience even with modest technical skills. )
(There is, however, more to [non-commercial] art than technical skill; a human artist of superior vision/passion/etc might still find a ready audience even with modest technical skills. )
In the future, it will hit a limit where it will not be able to separate generated AI imagery from real human imagery, essentially "polluting" the internet's pool of images. This may not seem like a technological limitation, but when you realise that this essentially creates a closed loop away from actual art, it will eventually fail to learn anything new and may even learn to reproduce the existing imperfections of AI imagery. I think that this will eventually make AI art useless or at the very least extremely hard to train beyond a certain point.
The possibility of a 'learning system' being drawn into a self annihilating death loop seems more likely if there are several 'AI' engines each feeding off the others' flawed output e.g. black NAZIs and English kings. [One would hope the developers of any given engin would take measures to prevent the simpler possitive feedback loop where the engine considered its own output as acceptable input.]
Screening training material to include only 'authentic natural person' material would be a solution, but would saddle the 'AI' developers with the same difficulty in definitively identifying 'AI' creations as faced by the rest of the net.
Screening training material to include only 'authentic natural person' material would be a solution, but would saddle the 'AI' developers with the same difficulty in definitively identifying 'AI' creations as faced by the rest of the net.
But does it not mean that other company like if i was other company and my competitor made better ai i would start feeding their ai bad images to send it into death loop and make my own avoid ai as much as possible??
That sort of scenario is oneof the things I would thnik those developing and maintaining any sort of expert-system/'AI' would attempt to gaurd against.
There are rumors that various players back in the days of the Cold War engaged in somthing similar by preparing false 'reseasrch' with the intent that it be stolen by others and cause the misdirection of resources.
Similarly, during WWII, I have come across stories of important looking (but nonfunctional) equipment being dropped under conditions where it might appear to be actual equipment from planes shot down with the intent to subvert NAZI research. (In at least one case, this appears to have included faked technical drawings for teh Norton bombsight.)
There are rumors that various players back in the days of the Cold War engaged in somthing similar by preparing false 'reseasrch' with the intent that it be stolen by others and cause the misdirection of resources.
Similarly, during WWII, I have come across stories of important looking (but nonfunctional) equipment being dropped under conditions where it might appear to be actual equipment from planes shot down with the intent to subvert NAZI research. (In at least one case, this appears to have included faked technical drawings for teh Norton bombsight.)
AI cannot control too much text, e.g. try putting in this prompt:
"A human with a deep fried head of lord farquaad from shrek, also make sure his eyes are squares and also there is a E on the image, it should also have a piano in there, and Mark Zuckenberg should be in the image" the AI will not include all needed things.
"A human with a deep fried head of lord farquaad from shrek, also make sure his eyes are squares and also there is a E on the image, it should also have a piano in there, and Mark Zuckenberg should be in the image" the AI will not include all needed things.
Interesting point… I'd not been considering the limitations of the front end/parser….
There is limit to how many thing it can understand and make but there is not for human??
I mean, if you include 900 details, an AI will use a picture with 100 of the details, but the human will add at least 300 of the details, on humans it depends on the skill too.
Makes sense okay thank you
It can't add text very well to images (eg signage and labels etc) or so ive found
yea they cant really do that yet
But other people in this forum say it can??? Now i am confused
Quite possibly different people are reporting results with different engines…
(Also, precisely how the prompt is written may be critical to getting teh desired result as the parsers ar still a far cry from an average human's abilities.)
(Also, precisely how the prompt is written may be critical to getting teh desired result as the parsers ar still a far cry from an average human's abilities.)
I hate AI it's stealing the jobs of artists authors and more ):
womp womp
Bro's personality is generated by AI 💀💀💀
But how is that a technical limit??
Sounds more like something that may trigger an equivalent of the butlerian jihad
It's more of a comment on the use of AI generated imagery to replace real artists within a system where your livelihood depends on the art you can create. The system and the corporations using it are the target issue, not necessarily the idea of AI generated imagery in general
For the sort of commercial art where there is no intent to make any sort of comment [beyond 'here is my widget in use'] I can see auto-generated images having advantages of speed and cost over human artists. (The history of moving animation from hand drawn acetate cells to computer assisted virtual cells may offer some insight into how the transition is likely to shape itself.)
This is what I mean - the brutal chase for cost efficiency technically powers progress, but at the expense of real people who need to live - AI models don't need to be paid minimum wage.
view more replies ( 1 )
Specific technical limitations depend on the engine: things like hands, teeth, skin texture/facial expression seem to be common problem areas… [Interestingly, these are also the bits of paintings/portraits that used to be done by the name/master artist (with apprentices etc filling in the bulk of the 'canvas').]
[I also strogly suspect that the level of tech floating about the web does NOT represent that actual current limits, the potential for propaganda generation is high enough that I expect the cutting edge will stay behind various governments security curtains until the fan made engines become effectively indistinguishable. ]
My guess is that (within 20-50 years) top of the line engines will be able to provide anything but the 'creative spark' [and eventually, POSSIBLY, that as well if general AI is ever achieved].
[I also strogly suspect that the level of tech floating about the web does NOT represent that actual current limits, the potential for propaganda generation is high enough that I expect the cutting edge will stay behind various governments security curtains until the fan made engines become effectively indistinguishable. ]
My guess is that (within 20-50 years) top of the line engines will be able to provide anything but the 'creative spark' [and eventually, POSSIBLY, that as well if general AI is ever achieved].
Okay so is like the internet using ai is not the best but goverment is hiding the best to keep us safe it sound like the internet origin hahaha???
Like internet was like military secret stuff and then it release and become wild west with only it technical limit but no ethical limit until more people get aces and influence so maybe ai is same story and we are looking at behinning of story now?
Like internet was like military secret stuff and then it release and become wild west with only it technical limit but no ethical limit until more people get aces and influence so maybe ai is same story and we are looking at behinning of story now?
Yes, various goverments are likely keeping back more advanced engines. (See the history of civilain GPS for a simple example of this sort of behavior.)
I'm not understanding your second paragraph.
In general, yes we are at the dawn of something that is likely to have widespread and important effects. [Note the latest 'cheepfakes' malarky with the US presidential administration accusing posters of faking content to impugn the physical and mental competency of Joe Biden.]
As AI reahes a point where whether an image is real, altered, or fake becomes something for debate mong competing experts, the utility of things like photographic evidence will be impaired.
Artistically, advances in AI will mean that technical competence in an art (e.g. painting) will dwindle in importance as someone with very modest tecnnical skills will be able to 'create' any work the can picture mentally via [probably iterative] description to an AI engine.
Other than expense, there wold also seem to be no barrier to hooking the AI to the necessary extentionals (e.g. a robbot arm and paintbrush) to allow teh production of physical works as well (e.g. an oil painting).
I do not see this as the death of art as this will simply shift the needed skill set for humans artists to produce top-of-the-line art. (At that, the old-skool [sic] ways will remain available, so it would seem likely that the net effect would be to expand the number of people able to produce technically impressive art. )
I'm not understanding your second paragraph.
In general, yes we are at the dawn of something that is likely to have widespread and important effects. [Note the latest 'cheepfakes' malarky with the US presidential administration accusing posters of faking content to impugn the physical and mental competency of Joe Biden.]
As AI reahes a point where whether an image is real, altered, or fake becomes something for debate mong competing experts, the utility of things like photographic evidence will be impaired.
Artistically, advances in AI will mean that technical competence in an art (e.g. painting) will dwindle in importance as someone with very modest tecnnical skills will be able to 'create' any work the can picture mentally via [probably iterative] description to an AI engine.
Other than expense, there wold also seem to be no barrier to hooking the AI to the necessary extentionals (e.g. a robbot arm and paintbrush) to allow teh production of physical works as well (e.g. an oil painting).
I do not see this as the death of art as this will simply shift the needed skill set for humans artists to produce top-of-the-line art. (At that, the old-skool [sic] ways will remain available, so it would seem likely that the net effect would be to expand the number of people able to produce technically impressive art. )
But will there be technical limit?? It sound like there will be no limit at all but that weird kind to me because everyone says everything has limit?
I don't see technical limits (oter than those on human art) — I do see various iimits based on expense (particularly once one move away from screen or printed images) and various legal impositions.
Tthere have been commercially available programs tat drive a cutter head to create bas-reliefs based on computer models q.v. https://www.cncsourced.com/cnc-routers/wood-cnc-machines-routers/
At present, none of these can produce corners that are truely sharp because the tools all rely on rotary cutters. (Whether the result from a 1/10 th mm diameter tool differs enough from a hand chisel or similar to be important is YMMV, but adding the ability of the CAM tool to use a chisel is a matter of cost.)
On a more fundatmetal level humans create art (or anything else) by manipulting either matter or information (or a combination).
Outside of the 'creative spark', I don't see any reason to think that there is any manipulation a human can perform that a machine cannot be made to copy. [Consider the signature machines that have been in use for quite a few years as another example; advanced models use actual pens to replicate a human signature.]
Tthere have been commercially available programs tat drive a cutter head to create bas-reliefs based on computer models q.v. https://www.cncsourced.com/cnc-routers/wood-cnc-machines-routers/
At present, none of these can produce corners that are truely sharp because the tools all rely on rotary cutters. (Whether the result from a 1/10 th mm diameter tool differs enough from a hand chisel or similar to be important is YMMV, but adding the ability of the CAM tool to use a chisel is a matter of cost.)
On a more fundatmetal level humans create art (or anything else) by manipulting either matter or information (or a combination).
Outside of the 'creative spark', I don't see any reason to think that there is any manipulation a human can perform that a machine cannot be made to copy. [Consider the signature machines that have been in use for quite a few years as another example; advanced models use actual pens to replicate a human signature.]
AI has no limits is scary because then if it wants to do anything it can it can create all paintings in world without flaw then that is scary to think because i could tell ai to make entire show for me and if ai has nonlimit it would produce a show as good as arcane within a day if ai has no limit
view more replies ( 1 )
It manly depends on what wepsite you are using, but mainly there is no limit to AI art. That meaning you COULD tell AI to make some really 'Weird stuff', but- For now, AI art is kinda cursed.
If there no limit to ai why cant it generate image or character that it not has data about??
Myave im not understand what you say but is that not a technical limit of it?
Myave im not understand what you say but is that not a technical limit of it?
Artificial intelligence will eventually reach the point where it can create any art, in any style, perfectly. Unless governments decide to do something, it will keep growing until it is practically undetectable. I won't say too much about my own opinions, but it is at least a bit worrying.
AI will reach that point if we keep feeding it, and it won't be able to create 'new' art, just reinterpretations of existing art. It cannot recreate an art style that it has never seen (if I were to develop an entirely new art style from scratch on pen and paper, and never upload images, the only way to get an AI to recreate it is to describe it in absolute, pain-staking detail, to the point that it might be easier to write a program to create it manually rather than train the AI.
I really need to start using Glaze and Nightshade on images I upload.
Anyway, right now we're also running up against a hardware limit. AI as it currently stands is effectively being 'brute-forced', using massive amounts of processing power that makes bitcoin mining look like a dude with a wood pickaxe compared to a Buildcraft Quarry at max speed (or some other mod-based automated mining tool). This is causing actual issues with power generation. If we want AI to take the next step, we'll likely end up needing to hook up the processing centers to a nuclear generator directly, or suffer regular blackouts due to power consumption.
I agree with your worry though, because it is becoming ever more difficult to spot when something is made with AI, and the mere possibility of video footage being AI-generated or AI-manipulated (AI-based editing tools rather than plain generation) casts doubt on ALL video footage by default.
I really need to start using Glaze and Nightshade on images I upload.
Anyway, right now we're also running up against a hardware limit. AI as it currently stands is effectively being 'brute-forced', using massive amounts of processing power that makes bitcoin mining look like a dude with a wood pickaxe compared to a Buildcraft Quarry at max speed (or some other mod-based automated mining tool). This is causing actual issues with power generation. If we want AI to take the next step, we'll likely end up needing to hook up the processing centers to a nuclear generator directly, or suffer regular blackouts due to power consumption.
I agree with your worry though, because it is becoming ever more difficult to spot when something is made with AI, and the mere possibility of video footage being AI-generated or AI-manipulated (AI-based editing tools rather than plain generation) casts doubt on ALL video footage by default.
Think Imma jump on the Glaze and Nightshade wagon haha
That why im also curious on technical limits of method but maybe yes there will come new ai with new method of generate image that will overcome current generating method issue?? Very scary yes but maybe also a intersting change in entire world like how it effect human but then im breaking my own rule of docus on technical limit hahaha!!
they can't draw snorkels right
For me there are no limits in the AI Art and it will continue to evolve little by little, and I think this is very good, the technology is just advancing, I also think that AI is the best technology ever created, and I can't stand anyone who is against it, it's something so wonderful, and something that really helps me create normal art too, when I started using Stable Diffusion, I really got addicted, there are much more dangerous things in the world, and no one says anything, so it doesn't make sense to be against AI.
I have a lot of more other opinions, but I will avoid saying here.
I have a lot of more other opinions, but I will avoid saying here.
Since you use it then may i ask is there something that the ai cant generate or where you have trouble getting good results no matter what you try?
Yes, some specific themes using a specific model was hard to come every tag that I put in the prompt, these specific themes is things that few people draw in the world, like, for example, with a model here, a certain very popular Pokémon was easy to generate well, now a less popular one was hard.
But it's what I noticed, and I just someone that just love AI, I still have to learn some things.
But it's what I noticed, and I just someone that just love AI, I still have to learn some things.
Assuming I understand the point correctly, these aren't technical limits; many current engines block attempts to create images based on certain prompts based on either legal or political reasons.
[Fundamentally, if an engine can create a clown, it can create J. Random Politician as a clown – not doing so is legal/politcal rather than technical.]
[Fundamentally, if an engine can create a clown, it can create J. Random Politician as a clown – not doing so is legal/politcal rather than technical.]
That is interesting thank you for sharing you knowledge iso so its like if there is no data of a specific thing it cannot be generated from specific angle
I wonder if it was train on like poses specfici it would be able to apply then but maybe it would not since poses are like guides and not actual final picture
I wonder if it was train on like poses specfici it would be able to apply then but maybe it would not since poses are like guides and not actual final picture
view more replies ( 1 )