Home
Creators
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register Favorites Keys Review DMs Logout
Community
Telegram ThePornDude
Home Creators Posts Import Register Favorites Logout

Generated Images 3: The Revengeance of the Menfolk (Patreon)

Published:
2022-10-10 20:10:41
Imported:
2026-06
Tags:
haven't archived this post yet. have a subscription? use the importer!

Content

All right guys. To end today's stream of new posts, let's have a couple more AI generated illustrations, this time focused on the guys.

Let's start it off by a winter casual Leo. While the version on the cover art looks cool,  this slightly softer version looks nice too.

A variation on the same theme. Imagine that this is the kind of style he has when he takes the girls out for a holiday date in town.

Last time I kind of ignored him in favor of a joke, but let's not forget about our resident protagonist. In his case, I was trying to get that quintessential "shounen protagonist hair" down pat, and this was as close as I could get.

Josh is a nice guy, just kind of over his head, so I tried to capture his softer side here.

On the other hand, when I tried to focus down on the "shounen protagonist aura", I've got this cocky Josh. He might've just won a basketball match or something...

Now, if we're at capturing different auras, I think this Mike turned out well. It got the gentle, kind of hapless tone right.

This one might do it even better, and I like this one because it gets the hair right; it's that fashionably disheveled look that takes hours of work at a hairdresser to appear completely natural.

I made a couple of Naorens, and in his case, the real problem was trying to pick out the ones to upload, as he had sooo many good generations.

I suppose the database the AI is working from is just naturally full of long haired pretty boys with glasses. It's a common archetype, after all.

Let's not leave the younger brother out either.

Very martial arts. Much wow.

The grand elder was infuriatingly hard to pin down, mainly because the long beard tag kept giving me characters that looked like Santa Claus.

While this one is a little angular, i think it kind of captures his intensity. You can practically hear him mutter, "You dare?!" on this one.

Speaking of old mean, while he's only a bit character, I figured I might as well try and make Peabody.

Since he's returning in the next arc, I figured I give Jaakobah an attempt as well. Don't as why he's underwater; the AI is weird like that sometimes.

But we can't have a full sausage-fest here, can we? Here, have an odago girl!

Here's an alternative take, with a weird qipao/kimono hybrid.

And last, but not least, every loli needs a chibi version, so please accept this Chibidango girl.

Files

Previews only

Comments

11037

Nice. I've messed with the same AI stuff myself in the last week. It makes wonderful faces, clothing, backgrounds, etc, based on whatever you type in. Very cool stuff. I do wish it could draw hands better. It really seems to struggle there, no matter what you type in. Still, it's impressive stuff.

Marius

First, I have to say that the pictures are cool, and I like that you're having fun. I have a feeling you don't know exactly how the AI picture generation works. As far as I know, it usually doesn't actually generate pictures at all -- at least, not in the way you a human does. It has a incredibly large database of pictures, usually stolen without the creator's consent. From there, it can mix the pictures and modify them to a certain extent to get what you want. Some modifications, such as colors, are easy. Others are hard. If the database doesn't have exactly what you specify, it will not be able to make it. For example, you can't make it generate a picture of a guy with an armband if it doesn't have any relevant pictures featuring an armband. The problem, at least morally, with AI generators, is that they steal from artists in two ways. First, they take their actual work, modifies it and hands it out, often for a profit. Second, since people can use AI generators, artists have a harder time getting commissions. Take a walk around twitter searching for artists complaining about AI generators, and you may get an impression of their feelings. Am I saying you should stop using AI generators? That's up to you. I just want you to have the full picture, which I had the impression you lacked. Now that you hopefully have it, you have to decide for yourself. I'm not judging.

11037
>>93334453

I understand the argument about art theft. Even if it's several steps removed from the original work, in an ideal world they would get paid something for its usage. Sure, maybe even make that the law. But I have much less support for the argument we should stop AI art entirely because people won't need human artists. Automation always hurts workers at first, and AI art will never be popular with artists, but automation has helped the world as a whole a great deal.

Vega
>>93332258

Have you tried entering super wacky information for the hands to see what happens?

11037
>>93332258

You don't need to type anything wacky. The seven fingered, melting together shapeshifter blob is the default for any prompt that suggests showing a hand. Any attempt to fix it, and it will usually mock you by making it eight fingers instead.

egathentale
>>93334453

Putting the slightly condescending tone aside, I have a feeling you are the one who isn't exactly clear on how stable diffusion works. In short, while the algorithm is trained using already existing artwork (in this case, a curated collection of images from Gelbooru), it doesn't just snip out and mix-and-match pieces of already existing pictures. Here's how it goes: first it generates a static noise seed, and through several iterations of sharpening, it slowly hones in on patterns in the noise based on the prompt given to it. If you tell the model you want a tall man with glasses, it will look at all the images in its training database that are tagged tall, man, and glasses, and slightly warps the image to make it closer to those. Repeat the process a couple of times, and it will result in a finished image that resembles a mixture of these images, but it wasn't copied from any of them. Don't take me wrong; this is disruptive tech that is going to shake up the niche the current deviantart crowd is holding onto at the moment, and it's going to suck for them, because this can be used to generate the kind of OC artwork they could charge money for. That's certainly going to be lost revenue for some, however this is not a new thing, nor is yours a novel argument. Luddites have been using the same arguments since the start of industrialization, where automation displaced low-lever grunt workers. The same is happening here: AI generation cuts into the revenue and displaces the kind of by-the-numbers commission artists who do piecemeal OC artwork, and it's going to really suck for them, but that's about it. You want artwork that's exactly up to your specifications? You still need an artist. You want complex compositions with multiple characters? Still needs an artist. You want perfect anatomy? Still artist. Lewds? You absolutely need an artist. In short, this tech simply automates the most basic, and at the same time most common type of art commissions on the net, and while that will definitely cut into the margins of the people who specialized in this field, it's just yet another unfortunate example of inevitable progress getting rid of the easy jobs.

Marius
>>93334453

I apologize. You are right, I don't know exactly how this works. My knowledge on the topic is limited to a single source, my daughter who is an artist and... suffice to say, has some issues with the AI business. Maybe I should go read myself up on it first and don't try to discuss topics I don't know enough about. I assume there is some kernel of truth to her arguments, but it is probably not as clear-cut as she presented it to me.

Marius
>>93334453

11037: I agree with your arguments; if the original artists is paid for their contribution, there is nothing inherently wrong with the AI generation. Well, some artists might get outcompeted, but there will always be some that need to modify their career with each new innovation.

egathentale
>>93334453

Oh, if she's working on simple character art commissions, it will certainly hurt her finances once this gets more streamlined and widely adopted. As I said, the AI models are really good at generating this type of artwork, and while it's never going to do as perfect of a job as a real artist, if you just want to make contextless visual aides for character designs (eg.: for forums, or D&D character sheets, or novel illustrations like I did here), they do a good enough job, and most of the time good enough is simply good enough. However, I don't think AI generated images are going to fully displace proper human-drawn artwork, for two reasons: scope, and specificity. As you see here, since this AI is specifically trained on anime artwork, it's really good at hair, faces, and clothes. However, it's abysmal at small details, like hands, and even once you get experienced with how the model works, it takes hundreds of tries to get something presentable. I've uploaded about, what? Fifty images? What you don't see are the literal thousands of failed iterations I had to slog through to get these. That's time and effort not everyone would be willing to put in, and if someone has enough disposable income to hire an artist to get things perfect instead of just "good enough", there's no reason for them to switch over to AI generation, as that would be a downgrade for them. As such, all I can do is wish the best of luck for your daughter, and hope that she'll still find her niche.

11037

A couple tips for anyone looking at NovelAI. First, placing brackets around a phrase seems to work as an intensifier. {{{{hat}}}} is more likely to generate a hat then just "hat" if you have lots of words. So if something isn't right or you want it to emphasize something, you can try that. Second, a very recent update now adds metadata to any picture you create, so if you upload it later, NovelAI can pick up whatever was used to generate it so you can generate it again. Naturally, that means other people can too and could see what words you used. Third, and I definitely for suuuure have never tried this, under the settings, the "NAI Diffusion Anime (Full)" has a greater selection of images to choose from. Also, in the Undesired Content section, the option to remove "Low Quality" effectively excludes certain tags from your results, including "nsfw". So if you set it to "None" instead ...perhaps a greater selection of content would be available to you.

Spicyice101

Specifically which AI was used to make these? I’ve yet to find a good anime character AI.