For Drawing References

For drawing references 

[@/ALEXbutalsoK on Twitter]

How to spot an AI grifter

OOOH BOY So, you know how it has become fashionable to generate pics by typing a prompt into an art theft program that consumed tons of hard work done by human artists, with no financial compensation? Some of you are capable of compassion and might wanna boycott people doing that, and especially companies that use such pics because they want to avoid paying artists (as if we haven't been starving already). However, AI has become pretty good at imitation, and not all people disclose using it, it's not enough to just look for a weird amount of fingers anymore. How to not fall for it? Here's a good example of how the art itself can be not too suspicious, but when you look at someones gallery as a whole, you can see that they're using an AI generator. DISCLAIMER! All those traits separately don't necessarily point to a person being an AI grifter, but their combination leaves me no doubt!

The most suspicious thing about this image is, I guess, the proportions being a bit wonky and the gold nuggets being placed chaotically, without the feeling of intent, but these are not clear signs. Even the hands seem fine! However, at this point I'm suspicious of any artstyle that's overly smooth and impressive at a first glance. Let's take a look at shubmedea's gallery on ArtStation! 1) First of all, I want to note the artstyle inconsistency. Of course, some people can draw in several styles, but the sheer amount of different ways of rendering raises suspicion

This style specifically is very popular among AI prompters, which made me more determined to investigate

2) The date of the first post. The account has only existed for 8 months, and it's already chuck-full of fancy looking fully rendered images. AI is a recent trend, so people who pump out pretty looking stuff and don't have any good artworks older than 1 year are extra suspicious.

3) The amount of posts per day. No human being is able to produce 8 pictures like this in one day. Of course, they can be posting a backlog of art from another platform, right? Let's give them some benefit of the doubt and keep looking!

4) The further down I scroll, the wonkier some images get on a closer inspection, and not like a human artist making a mistake. AI has its own specific ways of fucking up. It got better now, even at hands, but if you look at old images, you can notice things like these. The eye especially is not a mistake a person with the level of skill required for this shading would make

5) Some grifters just can't pick the best looking image out of the ones spat out by AI, so they post several versions of the same concept. Artists also can explore various compositions for the same idea, but I haven't seen it done at a stage further than a sketch on a scale of a full scene with a background. Usually the artist just does thumbnails for a scene like that, choses their favorite version and renders that. In case of the concept art for outfits, it's usually done on a copy-pasted standing base figure, not on figures in different poses featuring different backgrounds.

Here we can see 7 almost identical scenes rendered to completion, no artist would think it's worth their time to render something SEVEN TIMES. They'd rather choose another subject. Same with this girl, why render a very basic portrait like this two times, while slightly changing her facial features and the rendering style

Another example where you can just clearly imagine a person typing "gas mask anime girl hoodie cat ears neon colors close-up portrait [insert artist name here] style"

Note the variety of styles for each group of images. Even when spitting out similar gasmask hoodie catgirls, the AI does it with a different lineart thickness, some of them have solid cell shading and some have blending. The color choices are not consistent too, the one in the middle has a muted palette even with the bright-ish green on the hoodie, while the others are more neon. Why would a human artist make such choices when working on a series of portraits? Also, say hi to the wonky hand

6) Then I scrolled back far enough to encounter THE MOST typical early AI images, with abundance of details that carry no real information. They look pretty, but if you zoom in, nothing looks like it was structured by a human with intent, with thought behind how it works and why it's like this.

7) I understand that some artists are lazy, but most of us name our works publicly as something comprehensive, not like the three examples below. It looks like this person never named any of their "pieces", and never wrote any commentary on them. Again, as a standalone trait it doesn't mean this person isn't a real artist, but in the combination with the facts above, it shows their attitude to their "art". They treat it as nothing more than mass-produced pretty images to gain clout and possibly money.

8) Their page has no bio, lists no experience, but lists "concept art" as a skill! Also, no links to other social media, apparently it's too much work to post that many images on other socials. Very few artists keep all their art on just one platform.

I'm out of things to point out about this account, but I'll add things mentioned in another post: 9) Pay attention to teeth, especially in realistic styles and photos. AI fucks them up pretty consistently because it can't count how many there are supposed to be. The line between the front teeth is usually somewhat centered, it can be misplaced on some people irl, but realistic art rarely pictures people with dental anomalies unless it's a part of the design. Usually around 12 teeth are visible from the front (6 on the top, 6 on the bottom), but you can see more depending on how wide the grin/snarl is. Note the difference in shapes and sizes of teeth. Front incisors are usually larger than the second pair of incisors, then there are canines (pointy) and bicuspids (from the front look similar to canines).

10) unnaturally orderly/smooth fabric folds and skin wrinkles, sometimes too many of them 11) Inconsistency in patterns where consistency should be, take a look at the embroidery on this woman's sari, why would the pattern not be repetitive in a rendering style this realistic? (also what's up with the medallion's string being one-sided lol)

Thank you for reading this post until the end! It's devastating and stressful to see how people are already losing opportunities and jobs because now any rando can just scrape your entire gallery and generate something similar enough. It's not ok to profit off of the results of someone's hard labor without their consent! Please support human artists in these trying times, and don't engage with AI generated visual content, in 99% cases training data was acquired without the consent of artists, and it's no different than regular old art theft with extra steps. What we need is social pushback against art scrapers and law regulations that protect us from them. There already is a big class lawsuit filed against a major AI company, but its resolution will take time, and just one is not enough. If you want to help the cause, look for local initiatives, contact your lawmakers, sue if you have the money to, spread the info on the matter and boycott the companies that exploit stolen art. #team_human

Fantastic hands references by the website Hong14cafe.

Hong14cafe: Facebook | Forum

I appreciate a nice breakdown of this Trainwreck.