1 / 33
Oct 2019

So I found this28 free online AI tool, which puts two images through an algorithm to produce a result with the subject matter from the first image interpreted in the 'style' of the second image.

I tried it out with some random photos and some random art I had saved in a folder:

:doggo_shook: ...

And then, here's another one which makes anime girls:


... And here's one that makes abstract art.
https://arxiv.org/pdf/1706.07068.pdf2


Although the last one is academic (afaik so far), it's worth noting that the first two are both commercial products competing with real artists - you can have the AI generate custom artwork for you and buy it on the website if you want.

So, with that said... how long do you think it will take for artificial intelligence to make artistic skill largely obsolete in the marketplace?

  • 0-10 years
  • 10-20 years
  • A looooong time from now
  • Hah, it will never happen.

68voters

And what do you think will be the implications for art and artists?

Also if anyone knows of any other AI 'artists' out there, I'd love to see them :no_mouth:

  • created

    Oct '19
  • last reply

    Dec '21
  • 32

    replies

  • 3.8k

    views

  • 26

    users

  • 95

    likes

  • 8

    links

Just played around on the Waifu one after reading about it. It's... neat but obviously doesn't get down to some nitty gritty that people might want. For example, my friend wants a girl with glasses- after many refreshes, none of the girls that appeared had glasses.

It's not going to happen until you can get an AI with higher-order thinking skills. "Okay artbot, here's my passage/comic script with notes on the mood and tone for the scenes I want now start drawing." They can't that yet. You feed them a brief paragraph on how you want a character to look, they can't draw it.

i think we're forgetting that a human still had to create this code, and a human had to create the images supplied to these ais. to me, this seems more like a new form of art rather than something that will make all other forms obsolete.

I don't think they will ever make artists obsolete because there will always be people willing to buy art from humans, even if it's just to support artists or to say they bought something that was made differently from the AI-generated art... kinda like how people buy high fashion bags just because they're from a luxury brand lol

And just as a fun note... the waifu generator will also make things like... this

to be fair im pretty sure theres a junji ito comic with a character exactly like that

This takes that meme to a whole new level.

There was a video on Twitter about a guy asking his computer to make art. It was hilarious. I can't it but I thought it was funny.

There are a lot of details to art an AI cannot reproduce and I don't think it will be able to in near future or at all. Machines cannot really think outside the box. They do what they're programmed to and it would probably take years and years of coding to include every possible detail someone might want in their art.

Like in those links, AI can reproduce styles and colors, but cannot quite nail the mood, expression and all that stuff someone might be going for.

Also tried this waifu generator. It's pretty neat, but results sometimes get a little nightmarish. xD Tried the first one too to see if it can reproduce my style but it told me to wait a bit, so I don't know how it did yet.

To clarify, I'm not talking about artists becoming completely obsolete, but the point at which AI evolves enough to replace the majority of artists jobs and make it really difficult for humans to succeed professionally (I mean, it's already difficult but you know what I mean)

@stnmaren but what if ~5 people create an AI that is able to produce a volume and quality of work that puts thousands of artists out of business?

@OttoGruenwald oh yeah, I forgot that most of us here are concerned with comic art, not just illustration. That definitely seems like a much bigger challenge for AI to overcome. But what about, say, a program that takes a rough storyboard as input and outputs a finished page, or even just turns a detailed sketch into linework... or even just does the flats for you? Those seem a lot more conceivable but could still destroy a large amount of artist work.

@silacinnamon looool yeah... the waifu labs thing definitely isn't perfect. They usually look fine to a casual observer but upon close inspection of any image you're likely to find something that doesn't make sense

I think it could evolve into a genre all on its own because of the filters, but I don't think it'd replace most art unless someone makes a focused effort to feed an algorithm how to do it and not drive off of the uncanny valley like the Waifu and some of the anime generators. It's kinda like the attempts at AI books - it feels like a real person wrote them until Ron starts eating Hermione's family in that generated Harry Potter chapter, they'll always end up missing something unless they're honed to a diamond point, and maybe even then because most algorithms haven't goten there yet.

I tried the first one and it produced such a horrifying result from a photo of my husband and I and one of my drawings. XD I can't even show it here because it would cause sanity loss.

I could see it being useful in some ways, but since you do have to have a photo and a style image, I don't think it would work well since you would have to not only get those things but ensure you have the rights to use them too. I haven't checked to see what they say about copyright, but I wonder who the finalized image would belong to in this case. Even if it might not be a big deal now, someone might find a way to use it in an innovative way that is able to generate a lot of buzz and income and then the creators of the program may suddenly decide they deserve a piece of it.

I can say for the work I do, automation is very difficult. One of the things we have to do is background removal from images. Many users only have a JPG of their image and they need a transparent PNG. We have a tool that automatically does it with one click, however it really does not work well. It doesn't understand how to not be contiguous, can take out too much color, leaves a lot of artifacts, etc. So we end up having to work on those images regardless. Then there are poor quality images that we have to manually recreate or clean up.

We also have to vectorize images for production and anyone that has used Illustrator's Image Trace or Vector Magic will know that when it misses, it misses big time and some images just cannot auto-vectorize well because they are too low quality. Sometimes it doesn't even understand circles and text that is small don't even bother with. .-. Then it takes more effort to clean up the result than to just redraw it from scratch.

Our vectorized files also have to be a very specific way. An automation might be able to find some things... like if there are the wrong colors or a typo, but I don't think it would be able to do some of the specifics in the image itself.

So while an AI can make some illustrations, I think it would be difficult to do more specific tasks, working with low quality images, or generating content from nothing.

It's conceivable, but so is AI advancing to the point where it's basically human. As AI stands, they're good at certain kinds of thinking and bad at others. Before you see AI artists, you're likely to see centaur artists. https://thenextweb.com/contributors/2018/01/20/ai-will-turn-workforce-sweet-business-centaurs/4

Think some sort of device that draws what you think. "Okay I want this character to have body type A, and I want this character to wear a sweater." It would produce from a template but you'd have to put the composition together.

That's another good point. AI are bad at correcting. They aren't going to get "I want this character to look a little more fearsome, a little more menacing." Or "I want this character to look sad but not bawling-sad, like a quiet sadness."

AI replicates programmed patterns
Artists create the patterns

The problem with thinking like this is forgetting that people like other people, people like brands. People like something that is both different while also being reliable. The reason why people have favorite artist isn’t just the art style but what they draw- who they draw- not only that but often who the artist is. I’m sure these things will replace art that most people don’t care about- the pictures they hang up in their house that matches the curtains- but it will not replace personalized pieces. OC drawings, detailed landscapes, fantasy designs. These things are easy for a code to copy but almost impossible for a code to bring originality or “personality” maybe if one of these AIs have a personality and it’s own “brand” it will have fans that go to it, but at that point I think AIs with personality.... it’s gonna be a very different world by then.

As for my personal thoughts.... honestly? These feel less like “replacing artist” and very familiar to how most people react when they realize photoshop has filters and there are OC creation apps, they will be used maybe for reference and for fun but in the end it will not replace the “artist”

Here, I'll share my hideous creation with the tool xD

submitting a comic page as the "style" didn't seem to mesh very well LOL

it'll be a gimmick people get bored of quickly or implemented into art production to make things faster but still have human influence. Also it would be cool as fuck to just like sync up your brain to a computer and have the computer create your thoughts when it takes days just to figure out a character design.

IMO it won't. It will be complementary instead. Artist will use it to be "just another tool". Yes I know there was a machine-generated imagery winning an art contest7, but it's extremely tedious to make such kind of art.

Computer is not really that good for something that is not repetitive, is not structured in some way, and requires creativity. Machines can "learn" to understand human speech in English because the fact that English is semi-structured, and it does require gazillion data just to feed a program to understand us. For a language with so many colloquial form of speech, AI is still having a hard time, even google translate is still failed4

It's becoming a norm that big companies and mainstream media outlets really want us to believe that AI can replace everything and is solution to everything. But is it? I'd say it's half-truth half-lies.