The Image Generation Experience feat. PixAI
Friday, 23 August 2024
arrow_back_ios
arrow_forward_ios
It has been about a month since I did this AI art generation thing, so maybe I should talk a little about it. Just a little. *giggles*
Yes this is unprecedented, and many a times I have mentioned about why I still don't like the idea of AI taking over the world, but this is not the point, as I shall, in a nutshell, explain why you are better off NOT worrying about how much stress doing this is, and how everything you idolize and/or follow today is probably still better, though not safe, from this aggressive takeover.
Let's talk about why, yes?
(Nah, this doesn't count in the "What Otakus Do In Japan" series because they go see real things. *sob*)
WHY I STARTED
For why I decided to try an AI image generator out, I think it just occurred to me one day that I wanted to do this, out of curiosity, and out to fill in a blank. But the idea of doing so didn't happen in an instant though, this AI art thought. Actually, I started last year and thanks to Twitter/X recommended posts I have seen way more than I would want to, and some of them - often curated and then edited by the artists themselves - caught my attention, even if passing. But the explosion happened around the time when that "Engage Kill" mobile game was already ending service, and suddenly I have no mobile games to play other than "IDOLY PRIDE" as of now which is honestly saying, just an obligation right now. *silence* For months I've been looking at what I can do to fill in this space, from (yes) doing nothing on my free time at work lol just scrolling social media I guess how is Facebook a never-ending list of memes and successful lives of contacts lmao, to yeah spending more time on my blog right here which is also why the new version of this blog went on without much issue during July this year, but after that... Well, I don't know what else to do.
And that's why I had that thought of testing what AI image generation has to offer, and also, to offset my innate creativity (please, I was the only one who got an A on that art test in class when I was a kid *dabs*) apart from yes I write fanfics too why not take a look at those over here too if you are keen on my fictional life, I thought that the next best thing is to try out what others have been doing ever since the world exploded with AI last year and see how far I can take it. Not talking about rated stuff but I will talk about them here too lol hehe that's a thing by the way, so next up is to find which service will fit my into my life better.
I ended up with PixAI as mentioned on the title of the post mainly because PixAI runs on a credit systemย which means it has a free tier not just a trial period or full on paid service, and of course, they give you some of them for free when you login and publish some artwork. So basically you don't need a subscription or buy credit in order to generate images you want, and I will talk about the disadvantages when you don't, but otherwise, that wasn't the limiting factor, trust me. You can import and/or train your own models, but that's in another level which we will not talk about.
Going forward from here, I will only be mentioning from a normal user standpoint, using prompts on selected AI models to generate what you desire. Yes, your innermost desires... (No really, I mean it haha)
A NEW PERSPECTIVE
What I ended up getting out from this past month is rather spectacular but also rather cringe, in a sense. This really made me see a new perspective of art in general, and how real art drawn by real artists are so complete, and AI art can still have its flaws so super obvious, it'll only take me a few seconds to figure it out, and so will you, maybe. But more importantly ignoring the fact that this is literally stealing data from copyright owners who just want to share something with the Internet and are now in dozens of image generators out there I am so sorry how I know some of these styles because I know some of them, is that this is such a dedication to want to generate many, MANY images per day, and I tell you, it's taking over my life.
I capped myself to creating up to 5 images every day, but PixAI rewards you for up to 10 images, but that is if you have a lot of credit to burn and/or if you have a lot to time to burn, and I will tell you, one of them is more important than the other. *giggles* You could of course be like me and not burn credit to speed up the process (YUP) or find a model that doesn't use that much credit (there are literally 0 credit ones, but they are more... eh, "kinky", per se) (actually more like "literal", I will elaborate later) but then your sacrifice will be time. And that's why now I understand why they want to reward you for turning on your notifications, not really for the social feature since most of them just post on social media platforms to gain more "recognition" anyways, is because you can be informed when your images get generated get queued for up to an hour before they are generated. Evil, isn't it? *laughs*
I also, because I am basically a salaryman and I have a full-time job haha BRUH, do not have that much time to fine tune the prompts, wait for hours and then come back to them at work or even after work, because just like social media traffic, there are also certain times of day people will notice the stuff you publish, if quantity isn't what you aim for. Remember, they only reward you for publishing, they didn't cap the limit afterwards. *cringe* It sure made me more active at work, I'll tell you that much.
POINTERS IF YOU WANT TO TRY
So yeah, if you are that 12-year-old and why are you even here go away (lmao I'm sorry) and want to try things out for yourself and create images that you want, to better help us, here are some tips as to how you should go about doing this:
- Be Descriptive, Not Contextual
Most if not all the time, the AI models are somehow weaker at interpreting full sentences, so as seen from other works of "art", the prompts you write will have to be shorter and more descriptive in nature. So if certain terms don't work with the model, try using words to describe what they are. For example, if you know of the "jirai-kei" thing Japanese girls wear with the pink top and black dress and shoes with the twin tailsย and such? Then describe it, because chances are even if the prompt for the former is available, it won't do as well in some models compared to others. Selfies combined with outstretched limbs as your prompt for example, will also prevent those "mirror reflection" types of shots of the subject themselves, but that is up to the model.
Yeah so basically (haha) I spent the past month learning of new terms and how to describe them, thanks to ChatGPT (lol accomplice) I also learnt some Japanese along the way, because I also forced myself to type the captions to my published works in Japanese. Not that it changes anything I highly doubt people are searching by title there are tags that one should not search for too (#peeing) (haha) (please don't), but yeah, maybe it is helping in my vocabulary in some way or another, it will make your images hit what you want most of the time.
- Do Not Aim For The Impossible
As if this isn't obvious enough but I think I should emphasize again, do not, and I say DO NOT EVER seek for the perfect work of (AI) art. For starters, this is impossible. You aren't the artist, and there will be flaws in generating (mathematically, by the way) the perfect image for you, and because I know LLMs and how they run, this also applies to stuff that are impossible to the knowledge of the model mainly because the model hasn't been trained on stuff they never seen. So, don't try things like (and I have, by the way, tried) "idol bending down to thank fans" lol I tried too hard on this and gave up, to even "sitting on a sofa with arms spread out", not that assisted specific models (referred to as LORAs) will help out much on this aspect since there aren't any like this, but even if there is a thing for that, but the model can't do it on your first try, it's better to give up, really.
Which is why I went into this thinking that oh heck, it can generate whatever I want it to do (of course, safe for work) but ended up time and time being disappointed on what I can NOT create instead. What, I can't replicate the shot where Moeshi jumps out of the pool in her photobook?? Don't even try because I did. *laughs* I ended up with like, a woman in knee high water with splashes around her. LAME!!
Of course, there is the reference image feature where you can upload an image and they will try to replicate it in Anime character form I did try a few with them Suruga-ya bromides hehe, but do you really want to do that? I'm here to test its creativity, not give them more training material. *silence* Let's move on haha
- Give Up Early
Yes, this one. If you find something you like on the first time you generated something using your prompt, I can guarantee you 8 out of 10 times and trust me I went that far, the very first set of images you generated will have the results closest to what you want. It's a weird thing, honestly. That's how I got around to it too, and eventually be smart enough to not burn credit if I don't see something good I want above 3 tries, because chances are they won't get it anyways. Well, the model won't, so it's better to either move on, or tweak your prompts as mentioned above. Do not insist on what you want the AI to vision, because again, it's trained on images, not your own imagination, get it?
Then again, most people will just create AI art using Anime characters they know haha BORING
- Don't Copy, Experiment
There's a feature I like in PixAI which I use often, is that after I publish my art and wait a little while, you will be able to see related images (of course, AI-analysed) of what you have created. For me, the shorter the list the better because once again, it's showcasing my creativity in such things. Most people are just content with close-up shots, beautiful detailed eyes blah blah blah I generate those with my eyes closed, but if you came across what I have done for some of them? Yeah, you will be amazed at how I madeย two female space pilots sitting in a spaceship approaching Earth. *taps brain* I don't even have to lie on how well this works, the likes (though not much but that's a recommendation problem) tell the story.ย
Every single idea I created has a story to tell, and every time PixAI generated something for me, I have a new brain wave, like that time I thought a tube top dress would look better with a front zipper. *laughs* (Don't worry, the zipper was up) That's why it's fun to explore ideas sometimes ESPECIALLY if the prompts didn't go your way, or if you have new ideas halfway through. Which is why this brings me back to exercising my creativity, not creating half-a** works like many of them which is just girls posing in different outfits. Meh. *laughs* I like how I managed to make them pose as if they fell down because they broke their heels and look at me for helpย because THAT is what I call an Anime-worthy mishap. *dabs*
I also ended up experimenting with multiple girls in the same image, such as the one where it looks like girls in a locker room were preparing to go for swimming class (hell yeah) (of course, safe for work they already changed to their swimsuits) to even sisters having a barbeque party, stuff that many people don't dare to try because it's way more complicated to fine tune. Not gonna lie, I DID take more time on those especially that wefie with classmates after school oh boy that one was ROUGH haha it took me a couple of days and I'm still not pleased at the final product, but yeah, as much as I say that some things might be impossible, go for results with smaller tweaks to do, and explore ideas as you make them.
Another thing I notice many will not care even including me, is to not be satisfied with defaults. If you don't prompt them, they will just choose what they want: Blue eyes, long hair, focus on upper body shots... You get the idea. So if you want something away from the norm (which ironically brown eyes is a norm in real life), you will have to tell them.
- Iterate To Fix
And finally we have come to the hardest part of the equation: iteration. Yes, LLMs work that way I studied that in university, so suffice it to say that if you want to get the details right - maybe the food on the table looks a bit weird - then use their Edit feature and emphasize on what you want to change. It's not perfect and it doesn't work a lot of times especially if you crank up the similarity scale and then the image looks totally different now, but you have to keep tuning the generated images to get the results you would prefer. Which also brings us back to the opposite of this, where if you think one of the generated images is close to what you want, but no matter how many times you try the image won't budge, then it's time to try generating again or use other methods to generate a set of new images so you can explore other options.ย
This is itself a double-edged sword and I hate myself for the decisions I make with my limited credits. *laughs*
WHY DOUJIN ART IS BETTER
And come back to Earth, I will tell you now why ALL of this really doesn't matter much because real life artists are better, and here's why:
- Vivid Imagination
Real life artists' works are just INFINITELY better at this stage, before they are long time consumed by the AI art models and then creativity is no more lol, meanwhile though, if you have seen how some of them are dedicated to their works like tetto-sensei who literally drew a "jirai-kei" girl fixing a lightbulb and went to vote in the local elections (lmao) you would understand why maybe "making" your own images isn't that interesting and/or fun after all, if they can't create what you exactly want. Remember, AI art generators aren't your brain, so you can't do that better than... well, taking art classes on YouTube (lol) and creating them with your own hands. And then get them stolen by AI companies haha okay maybe not that part, but you get the point.
AND DON'T YOU DARE TELL ME TO USE THEIR IDEAS TO GENERATE IMAGES LIKE A DEGENERATE AND GIVE THEM TRAINING DATA BECAUSE NO (haha)
- Attention To Detail
Yeah yeah, I know what you are thinking, there are detail tweak LORAs specifically designed to make things look more precise and detailed, but heck once again, if you don't have a brush or pen (or Apple Pencil lol) then you don't have that power to pinpoint what you make, and turn it to whatever you want, and that applies especially to certain XL models in PixAI. "Perfect fingers"? You better hope your viewers don't zoom in too much. (Her right hand is kinda...) "Full body"? I hope you are ready to sacrifice quality on face details. So you see, this is not what AI art will do or would even aim to create, which is why every time I write my prompts I have to include "masterpiece" into it or add in negative prompts, so it can filter the infinite number of sh*t artworks they created in order to reach a final result you won't puke on.ย
And trust me, I saw stuff that never should've seen the light of day. *cringe*
- Seeking Perfection
An upside to real world artists is that, unlike their AI "counterparts", these guys will strive to create the perfect image. I mean, I know there are perfect AI images, but that's also because they are touched up by the artists themselves that will create that now combination of quality in quantity. And that's likely how things will move forward for the industries that have to live with it, working along with AI art generators. Also, just like everything out there now, we would expect if humans can do 99%, that AI should be 100%, but that is CLEARLY not true if I have to put "masterpiece" on my prompts. Or was it that it can't understand perfection? We will never know.
- Profession Requires Control
Over the past month, the most frustrating thing I have experienced ample times is that, among other things, generative AI is hard to control. It's either the engineers were smart enough to make it work the way you want, or then you might as well do it yourself, train out a model and such. And that's where AI has its limitations, but that's why it is now complemented by real life artists and their own touches, as mentioned in my previous point. This also points us back to the idea of not trying for anything in your brain, because in super low light situations where I tried someone walking in a dark room basically hand against the wall because it's that dark ended up being just someone walking in the corridors with dim light, somehow it's hard to create unless you force it into some algorithm which will somewhat take way more effort, to be frank.
To have such control means one's artwork should gear towards what they want, consistently. Heck, I can't even create the same girl again for another scene, like, right now! How do you think I can be consistent about my work in a month?
- Low Effort Community
Not that I have done much in terms of engagement, but what really brings to my attention is how much, HOW MUCH TRASH there is online even inside the PixAI website. User submissions are often duplicated likely because they are testing or they want to rake in the rewards, but also more than likely, is that they don't care. Many don't even have the eye for detail, and I wouldn't even consider myself one sometimes considering I regret what I published fast enough that I can't take it down since someone still liked it. Yeah, it's that vexing. And that's what is polluting social media right now and creating an influx of weird stuff that will no doubt become AI's own poison (as with all things in other industries running online right now) because this freedom of generating whatever you want will end up making specialization of such artwork more scarce, but less noticeable even if the next big thing is around the corner.
I don't consider my work top notch but my ideas are novel, and that is what's wrong with the community who doesn't give a f**k. There, I said it, happy now? *laughs* Like, who in the right mind will make an Anime girl angry showing off her bikini? Or getting caught byย a cheeky witch? Or even just sad that summer's almost over? Think about it.
RATED CONTENT
As a side note, and that's why I don't want to talk about this too much, even though PixAI has a strict moderation and AI-generated scanner for published artworks, I have also been dinged half a dozen times throughout the month for posting stuff.. that honestly saying isn't illegal by any means, not even for minors. Like, what's wrong with a photo of looking at Anime girls from their backs on the beach? That much rear can't get pass the YABAI filter? *shrugs* I haven't even tried prompting for profanity but damn some of the models will show you nudity if you don't provide them some reference for clothing, for example. *giggles* You get the idea, I won't even dare to publish those, but if the difference is between donning a naked towel submerged in the hot springsย and NOT having that on (image link not available haha) will get me cancelled, then why isn't Marinka's photobook where she was in a bath rated content-
Haha I think you get it, and if you want to generate images to jerk off to for your own pleasure, you know where to go. *laughs*
To conclude, and I can go on forever but I won't because I know what you are here for, let me showcase (hehe) (just a bit more) of what I have done in the past 30 days (not including today) so far!! *laughs* If you want the latest stuff I am still active but will be less than the past month since I just spent a couple of hours writing this today after work lol, but yeah, with an outstanding 150 images created just so I can brag about how I created 5 images per day on average haha yesterday I had to churn out 8 of them actually since the first few days were just playing around, here are some of the highlights!
One of the first few images I tried, did only a few iterations but looked surprisingly good overall (Yes, I want to see this on my bed IRL haha) |
If you still want to see more, feel free to visit my PixAI profile. No rated content though, you can't see those anyways without an account. *giggles* Heck, I'm pretty sure you can't even see it, flagged is just condemned for life, dude.
I don't know for how long I will do this, honestly. I'm getting rather frustrated to even bored of the iteration already, but as one of my followers (lmao) wrote on their profile, is that they hate themselves for not having the talent to draw. Yeah, because if I do, I won't even bother to dabble in this cr*p, honestly. *giggles* Like, right? Need I say more? On the other hand, I am still enjoying this play on creativity where I can offset on days I have too many ideas on my mind, so that isn't a bad thing and it's still fun, so unless I find a new mobile game to play or find myself a new hobby *looks at Alternative Girls 8th anniversary livestream*, this might be the new thing I continue to do on a daily basis.
*silence*
I wish I was born an artist.
P.S. Will update with as many references of images as I can over time, my arms are stiff from writing this essay which is why I don't even want to do this even if it is to make a statement.
You're welcome.