About AI covers.

[RepresentingWrath]

Recent profile post by Dieter about AI covers has developed into a brief conversation with a friend. This conversation got me thinking of one thing. Why are all AI covers look so similar? You have noticed those covers as well, right? No offence meant to authors of these stories, I just grabbed the most recent stories on SH with covers that suits the description.

No, it's not about limitations of AI, AI obviously isn't omnipotent, but it can produce something crazy if you think outside of the box. This statement begets a question, are people just lazy, or is it really a lack of creativity?

I am lucky, I have a friend who knows photoshop so I can ask him to edit this or that. However, even without him, I would edit covers on my own, either with Canva or other sites. Aside from editing, all those covers just look... bland... I know they work. I'm not here to diss people using AI; you can even dismiss this half-rant by saying, "If it ain't broke, don't fix it." I know this, but I still want to hear your opinions. Why was I able to do something more unique two years ago, with a free AI gen site, than people now? Is it really as simple as laziness or is it AI models problems? Maybe the problem lies with people trying to substitue real artists with AI instead of fully utilizing AI? They try to create 'the character in their head' but end up with those generic-looking art pieces. Share your thoughts.

P.S. I won't share my AI pics cause I'm lazy, so you have to 'trust me bro'.

 

[Representing_Tromba]

It depends on the art the AI model is trained with. Sometimes they all look the same while others look a bit different. Though they are limited to that style, and so, very bland.

 

[RepresentingWrath]

It depends on the art the AI model is trained with. Sometimes they all look the same while others look a bit different. Though they are limited to that style, and so, very bland.

Yes and no. I believe you can still achieve something different. At the very least you can try doing a different pose. Even if AI shits itself, you can hide scuffed parts with title.

 

[Tempokai]

GenAI is designed to give you a median output. That's the fact everyone forgets. Median, in this case, is how often that thing happens according the training data, and generating the average. Because that average is, well, average, it differs on the input. Some put ungodly amounts of text to make the precise image they want, while some just write two lines of prompt, and rely on the randomness (aka training data of the model) to make the image. Are you seeing where it is going?

Because the user doesn't use a lot of time to make an image, and the cost of GenAI is cheap perceived computation wise, the user has all the incentive to grab the one of the first four images. The human brain is wired in that way, if the brain sees great value in short time, it defaults into "good, grab it now!" mentality. The perceived value is great at that moment, and most people think right here and right now, because the thing they wanted from GenAI can make it right there without any loss on the user. Multiply it by hundreds, thousands, millions, and you can see the problem. It becomes average.

Due to average user not being well versed with art (even if they consume art everyday), they don't know that they are average, and average anime art of their "protagonist" in the portrait, in fantasy clothes, without lineart is so average that some users notice that pattern. Because they don't think in systems, don't think hard about the cover image, they just leave it as is, because at that moment when they generated what they wanted from GenAI, their mind was satisfied with that current image. That's called "not knowing what you don't know", or in simple words, ignorance.

If you think in averages in the AI cover making, it's too visible that ordinary users just don't care about that. They see their waifu or husbando, and just leave at that. If they know some editing software, they'll add some text, but that's their limit. They don't know how the pose works, how color palette of the image works, how portraits work and why they have passive pose instead of active one most of the time, they just see a pretty picture. When ignorance consumes the creator, they'll become average, and average in art is DEATH.

 

[RepresentingWrath]

GenAI is designed to give you a median output. That's the fact everyone forgets. Median, in this case, is how often that thing happens according the training data, and generating the average. Because that average is, well, average, it differs on the input. Some put ungodly amounts of text to make the precise image they want, while some just write two lines of prompt, and rely on the randomness (aka training data of the model) to make the image. Are you seeing where it is going?

Because the user doesn't use a lot of time to make an image, and the cost of GenAI is cheap perceived computation wise, the user has all the incentive to grab the one of the first four images. The human brain is wired in that way, if the brain sees great value in short time, it defaults into "good, grab it now!" mentality. The perceived value is great at that moment, and most people think right here and right now, because the thing they wanted from GenAI can make it right there without any loss on the user. Multiply it by hundreds, thousands, millions, and you can see the problem. It becomes average.

Due to average user not being well versed with art (even if they consume art everyday), they don't know that they are average, and average anime art of their "protagonist" in the portrait, in fantasy clothes, without lineart is so average that some users notice that pattern. Because they don't think in systems, don't think hard about the cover image, they just leave it as is, because at that moment when they generated what they wanted from GenAI, their mind was satisfied with that current image. That's called "not knowing what you don't know", or in simple words, ignorance.

If you think in averages in the AI cover making, it's too visible that ordinary users just don't care about that. They see their waifu or husbando, and just leave at that. If they know some editing software, they'll add some text, but that's their limit. They don't know how the pose works, how color palette of the image works, how portraits work and why they have passive pose instead of active one most of the time, they just see a pretty picture. When ignorance consumes the creator, they'll become average, and average in art is DEATH.

Thank you for your input! This is an interesting take.

 

[RepresentingPride]

Yes and no. I believe you can still achieve something different. At the very least you can try doing a different pose. Even if AI shits itself, you can hide scuffed parts with title.

it all depend of the prompt to be honest and how you use it to have what you want, if you look for generic, you will get generic. Here some AI (I didn't said there not some generic in it, but at least the style differ):

Spoiler: Images

 

[RepresentingWrath]

it all depend of the prompt to be honest and how you use it to have what you want, if you look for generic, you will get generic. Here some AI (I didn't said there not some generic in it, but at least the style differ):

Spoiler: Images

View attachment 39295View attachment 39296View attachment 39297View attachment 39298View attachment 39299View attachment 39300View attachment 39301View attachment 39302

Even if artstyles aren't super diverse in your examples, different poses and colors make those pics a lot less generic.

 

[LilRora]

I believe this is because good human-made covers have a certain symbolism and purpose, or are spontaneous. There's a kind of idea behind art, emphasis on something that is supposed to attract attention within a narrow style. This may be a little muddled with low-quality covers though.

AI covers, though they're improving, usually have none of that. It's just a static piece of art that may look solid at first glance, but doesn't have any real depth behind it. For this reason, AI is good for creating something that is supposed to be big, complex, intricate, unclear, but it starts to struggle with simplicity and with style.

 

[RepresentingPride]

Even if artstyles aren't super diverse in your examples, different poses and colors make those pics a lot less generic.

and that some junk product

 

[RepresentingWrath]

I believe this is because good human-made covers have a certain symbolism and purpose, or are spontaneous. There's a kind of idea behind art, emphasis on something that is supposed to attract attention within a narrow style. This may be a little muddled with low-quality covers though.

AI covers, though they're improving, usually have none of that. It's just a static piece of art that may look solid at first glance, but doesn't have any real depth behind it. For this reason, AI is good for creating something that is supposed to be big, complex, intricate, unclear, but it starts to struggle with simplicity and with style.

That's why I think people should go crazy with AI instead of trying to imitate a real artist and 'draw' a simple portrait with AI to use as a cover.

 

[Sola-sama]

Can't blame AI users,, they are cheap.

If it were up to me, I would commission Wlop, or maybe Thorkel40k for my artworks but am piss poor, so AI is the best alternative.

 

[RepresentingWrath]

Can't blame AI users,, they are cheap.

If it were up to me, I would commission Wlop, or maybe Thorkel40k for my artworks but am piss poor, so AI is the best alternative.

Not Wlop brother! They will think it's AI. ?

 

[JayMark]

The cheap AI I use always has a flaw. I have to scan through hundreds of images and when I find something mildly useable I have to check it for being cursed. If it's not overly cursed and approaches the idea I want, I'll file it and continue generating.

I'd much rather have a human made cover that I made myself.

 

[RepresentingWrath]

The cheap AI I use always has a flaw. I have to scan through hundreds of images and when I find something mildly useable I have to check it for being cursed. If it's not overly cursed and approaches the idea I want, I'll file it and continue generating.

I'd much rather have a human made cover that I made myself.

That's not the point of the thread. UOOOOHHHH! ? ? ? ?

 

[JayMark]

That's not the point of the thread. UOOOOHHHH! ? ? ? ?

WHAT DO YOU WANT FROM ME!? Moo.

 

[ChubbyLiv]

Can't blame AI users,, they are cheap.

If it were up to me, I would commission Wlop, or maybe Thorkel40k for my artworks but am piss poor, so AI is the best alternative.

WLOP? That would be like a 10k commission, no?

 

[Sola-sama]

WLOP? That would be like a 10k commission, no?

Honestly, I don't know his/her rate,, but I've been a fan of their artworks ever since I had an AK.
by that I meant ever since when I was able to walk.

 

[Tyranomaster]

My AI art thread gives a little bit of guidance on this, and some people here have touched on it. In essence, if you just prompt AI with like "man in business suit with chinese dragon encircling him", the only information the AI has is that. You haven't provided specified styles or techniques (which almost every modern ai image generator knows), so you get an amalgamation of the average.

A few months ago people were all doing "in the style of studio Ghibli" images. There is a lot of stylization information you can provide to make things look in a particular way. If you are doing gen on somewhere like civitai, you could even add particular artist loras into the image to get particular artstyles.

In short, they all look the same because no one bothers to stylize. Hell, even I don't bother most of the time because I'm only looking to cross the minimum bar with my cover art. I might try some with my next few, just to mix it up.

 

[HisDivineShadow]

I agree. I use AI for drawing, not only for covers, and it takes a lot of time, because of all the detailing. Right now, I have two covers . Both were made with AI. One went through several iterations and redrafts; the other was generated on Leonardo in just one go. Can you guess which one took only two minutes? But when I replaced that simple cover with the more polished, detailed one, I saw a major drop in views. So now I’m not even sure anymore: do people really want detail, or just a bold, catchy image that grabs attention?
And yeh... I really don’t feel like paying a hundred or two for a cover to a story that’s not even finished yet and hasn’t earned a single cent.

 

[Dountnothere]

GenAI is designed to give you a median output. That's the fact everyone forgets. Median, in this case, is how often that thing happens according the training data, and generating the average. Because that average is, well, average, it differs on the input. Some put ungodly amounts of text to make the precise image they want, while some just write two lines of prompt, and rely on the randomness (aka training data of the model) to make the image. Are you seeing where it is going?

Because the user doesn't use a lot of time to make an image, and the cost of GenAI is cheap perceived computation wise, the user has all the incentive to grab the one of the first four images. The human brain is wired in that way, if the brain sees great value in short time, it defaults into "good, grab it now!" mentality. The perceived value is great at that moment, and most people think right here and right now, because the thing they wanted from GenAI can make it right there without any loss on the user. Multiply it by hundreds, thousands, millions, and you can see the problem. It becomes average.

Due to average user not being well versed with art (even if they consume art everyday), they don't know that they are average, and average anime art of their "protagonist" in the portrait, in fantasy clothes, without lineart is so average that some users notice that pattern. Because they don't think in systems, don't think hard about the cover image, they just leave it as is, because at that moment when they generated what they wanted from GenAI, their mind was satisfied with that current image. That's called "not knowing what you don't know", or in simple words, ignorance.

If you think in averages in the AI cover making, it's too visible that ordinary users just don't care about that. They see their waifu or husbando, and just leave at that. If they know some editing software, they'll add some text, but that's their limit. They don't know how the pose works, how color palette of the image works, how portraits work and why they have passive pose instead of active one most of the time, they just see a pretty picture. When ignorance consumes the creator, they'll become average, and average in art is DEATH.

I was trying to explain this to some people and this exactly what I wanted to say.

Also I don‘t think most or average ppl are all that much about Ai. Since they wouldn’t be all that knowable about art.

Before AI, I used say this to peers, ”don’t worry about how ur art looks or about ur skills.”

This because the general person they will still look at your stuff with amazement or like it anyways.

It’s only when you’re sharing art with fellow peers or art friends that they might look at it differently.