[August 2022] Creative AI is changing visual culture | Workshop Talk
HTML-код
- Опубликовано: 19 май 2024
- I gave this talk at a live workshop hosted by Han Xiao of Jina AI jina.ai/ and the team behind Discoart github.com/jina-ai/discoart. It was a fantastic and engaged audience with a great Q&A and roundtable. I only regret not being able to spend more time in Vienna. Thanks to Joseph Hillenbrand for running these prompts through DALL-E 2.
Check out Jiawen Chen's illustration series here:
jiawenchen.art/jiawenchen-ill...
And the packaging design from OMSE featured here:
thedieline.com/blog/2022/6/23...
0:00 Art vs. image making. Asking the right questions.
5:34 The design process overview
9:00 Challenge: AI vs Human. Reverse engineering creative outcomes.
20:59 A short history of disruptive change in media
24:51 Opportunities and challenges. Utopia vs dystopia.
![[June 2022] Will AI really make graphic designers obsolete?](http://i.ytimg.com/vi/LMT-RO-_TbA/mqdefault.jpg)
The state of AI art makes me feel very sad. It’s quite dystopian in my opinion as most of the people who use it will not use it as a tool but a replacement. The fact that nowadays it seems to be collapsing in on itself truly brings me great joy.
One problem I have with the idea of synthetic, single use, reaction gifs, is that at least half the time the reaction gifs we use are already well known.
The Gene Wilder Willy Wonka gif, or that Michael Jackson popcorn gif, are used often, by lots of people, to portray the same idea every single time.
People would want to use the same reaction they saw someone else use. Beating the same dead horse is a big part of internet meme culture, and I can't see the desire to repeat memes being overridden by the ease of AI generation.
I suppose it comes down to "what gets my point across faster?" Finding the meme everyone knows and recognises instantly, or generating something bespoke for that situation?
I feel like memes are usually going to win out here.
I had the same thought but realized it would also open up the space to generate your own variations on existing meme formats much faster. Could even end up... putting yourself in well-known reaction images in a much more creepier high-tech way than the face template stickers we have on Snapchat (and probably other platforms) right now. Don't think it'd replace meme culture but it could definitely change how people interact with it and how memes spread.
The beauty of it is mixing popular reaction memes with other images. Meme remixes. Which is already a thing but AI is going to put meme reviewers in a pickle since memes trends run 10 faster than current media news cycles.
watching this 9 months later... what is amazing is how good AI got in so little time
I’ve never given a conference talk in shorts! Well not an in-person one anyway. Dammit now I’m envious.
Haha, then have you even truly lived, Rohin? It was 36 degrees in Vienna that day and I packed light! 😄
Thank you for uploading this workshop to RUclips. Very informative and thought provoking, great stuff!
there’s such an interesting crossover with linguistics here!
you’ll often see linguists talking about how “referents” in a clause are usually extremely ambiguous without (and sometimes even with!) context. if you look at the maze example, “a woman navigating a maze shaped like a human head” (paraphrased), you’ll notice the adjective phrase “shaped like a human head” isn’t specifically about the woman OR the maze.
it’s obvious to us, as people who have heard language and it’s infinite complexities through a massive organic computer our whole lives, because we’re comfortable applying certain boundaries (ie. it’s more likely to us that the MAZE is shaped like a head than the woman is, because we understand the implications of a woman and what is means for a woman to be in a maze). the ai, however, is trained on a comparably tiny number of scenarios taking into account the detail (or lack thereof) it strings out of them, and it’s unknown to the user what sentence structures the ai can comfortably and correctly interpret.
unpacking that example further, you can see that most of the ais take the word plant (describing a detail in the left corner) and apply it to the entire artwork; maze and all. with its rudimentary understanding of language composition, you can see again how the ai struggles to isolate descriptors to discrete parts of the sentence.
it would be interesting to see how these text to art generators develop in the future to accommodate these pitfalls. i could easily see the field developing some standard shorthand for the user where each new sentence is interpreted in isolation, and a certain format has to be used as to eliminate ambiguity. something along the lines of:
[article/freq.] [descriptors] [subject a] [verb] [preposition] [article/freq.] [descriptors] [object]
so your prompt input becomes:
[NORTHEAST, MIDGROUND]: [A] [] [woman] [walking] [in] [a] [large, stone, human shaped] [maze].
[SOUTHEAST, FOREGROUND]: [Some] [small, leafy, tropical] [foliage].
STYLE: [flat illustration, alegria, facebook]
what this does do is a) make the sentence easier for the ai to interpret, which makes it easier to get what you asked for, but also b) makes the technology less user-friendly and c) somewhat removes the artistry of creation. although i suppose this is kind of what ai art does regardless, so it’s just pulling back the curtain somewhat.
the sentence structure limitations actually have quite a few echos of child language acquisition, so perhaps it’s better to think of the prompts and the ai as children learning to read lmao. you start basic and simple, until they eventually learn how the world works a little better, and then you can start introducing more complex clause arrangements and abstract concepts that rely a bit more on learned context.
it’s wonderful to see how two fields as wildly different as synthetic language comprehension and graphic design get brought together like this. the world is so damn interesting!
Excellent talk. Everyone's knee jerk expectation is that artists/designers will be out of a job next week. It's understandable - the art is quite jarring to see for the first time. But it's not the full story. It's a tool like anything else, which means it can't replace the entire pipeline of artist-client interactions. I used DALLE-2 myself and this quickly becomes obvious. It's just tricky to guide the tool into exactly what you had in mind..
I think it's similar to people thinking that squarespace and general web builders will displace web developers. That's clearly not the case.
What seems to happen instead is that highly automatable work becomes easier to get through. Skilled labor is still necessary for more creative tasks. We just get rid of the really boring stuff (stock photos in this case)
The only use these AI image creators will have is for inspiration. So you give it an idea and it spits out something that you can then use as a basis for your own piece of art. But not copying it. Just using what's good and getting rid of what's not.
The novelty of seeing "whacky" things like "giant octopus eating Gordon ramsay lololol" will wear off very very quickly. I mean, it already has. All the possible jokes have been told already, every new one is just a variation on those.
@@duffman18
nah
They also have use in doing the much more menial tasks of making art. In fact, it's already been used for that application for a long time now. Photoshop has had their content-aware tools for a long time now.
Well I'm a low end budget storyboard artist. I've been told point blank that they don't need my services or my peers either because they can "get something usable out of AI". A lot of people already feeling the the pinch. I respectfully disagree. People are in fact getting discommoded.
@@GregoryGiordano Really sorry to hear that. If anything, the best is perhaps to not pay attention to those claiming that "it's just a tool", because these people are always going to be the ones who never had any desire to work on art before, and who are jumping on the AI train in hope of making a quick buck.
I hope you'll land on your feet - your work looks great.
Hi Linus! I always appreciate your level of analysis. An art and design thinker on the cutting edge--testing new technologies with a rich background in traditional approaches. Glad I found your channel. Your videos make me want to think more intentionally about my visual choices in the art that I make.
I really appreciate you putting the subtitles in all your vids!
I find it frustrating that the captions substitute "[ne'er-do-well]" for the spoken "dickhead" - those of us who use captions receive a sanitised (and not correct!) alternative to those who can access the talk without CCs. Otherwise this talk was super interesting though, really enjoyed it
Ha! I usually aster*sk out the expletives, but google's auto transcription totally censored it, changed it to: [__] - ne'er-do-well was just a whimsical substitution, but I see your take! I'll amend that.
Great talk! I enjoyed the discussion of what new uses might be enabled - starting about a year ago my group chats have included quick AI-generated images referencing in-jokes and capturing the same kinds of real-time reactions as GIFs. One of many weird things enabled by this new tech :)
I've been trying to get decent typography results from AI but nothing usable from the bet go. A lot of great ideas but I don't think AI can handle anything with precision yet.
Thank you so much for a fascinating talk!
Beautiful presentation, with beautiful typefaces, that's art too
i love the chumbox lol
i feel like past disruptions could include the calculator (will it replace mathematicians?), the camera (will it replace portrait/landscape paintings?), and digital painting (will it replace traditional?)
these aren't perfect analogies to the potential impact of AI images, but i think they contain partial visions of it
Shared to all my designer friends and business pals.
I think on our side, we're all excited to see where things will go with AI and Design.
Really enjoyed your talk! AI isn't something I normally consider/think about as a designer but I think I need too more! found it really fascinating, thank you :)
Thanks for sharing 😃🙏
Great talk, definitely here for all the new ways we're gonna be able to use AI and implement it into our design routines. Boy would I love a break from compositing an image for 5 hours and get the result in a few prompts!
Unrelated to your comment but I love your Night in the Woods profile picture 🧡
Fantastic talk Linus!!
Very cool talk!
7:25 Why does the letter e drop to a separate line in the text bubbles? When you show the slide intself on screen everything's fine.
Eagle eye! For some reason the text boxes in that section changed on the laptop at the venue. The cutaway was taken from playing back on my personal machine after the fact. Still not sure exactly what happened there!
Excellent Talk!
Fascinating topic Linus, really enjoyed your presentation! I am quite new to this whole thing, so excuse my ignorance, but I thought you, or maybe another viewer can shed some light on this for me:
How come, most of these AI image generators fall short when you ask them to include written words/typography in the artwork? Based on the core concept of 'text to image', it's like AI can read but it cannot write...? Is that a deliberate restriction from the programmers' side?
Hi linus, I really loved the animations in the presentation. Can you share what tools you used to make it?
Actually believe it or not - it's Canva! First time using it in years, and actually found the simple interface very helpful in putting together a clean and minimal presentation. partner.canva.com/timesnewboman
@@LinusBoman OMG, what??? 😳. Lol. So it's really the artist that matters, not the tool
Hey Linus, have you ever considered doing a typography -online-class/course or anything like that?
nice font selection damn
Looking at your design process flow chart and also the previous video... I feel like that's what might become of the models. Instead of you figuring out a prompt by changing keybowrds. You might simply get into a bit of a conversation and offer feedback (text, masks, images, arrows, annotation) back to the model.
Langauges models can be fined tuned and prompted to solve problems with reasoning, or explain jokes. So it isn't far fetched to have the model generate images but also explain some ideas behind it. Perhaps even deconstruct an image into parts, layers and provide conceptual sliders.
Make the model mimick the experience for an actual customer to gain additional context and produce better results.
But I will now continue to watch your talk.
The idea of "single us images" exist in like Snapchat, Stories, BeReal, but I doubt instant selfies will be replaced by generated images. Perhaps modifying images, which you currently see with the variety of filters, but higher fidelity.
Great talk! Very insightful and thought-provoking. I think you nailed at formulating the possible future of this technology.
great talk! i told my roommate about your channel and he didn't know what typeface was but he DID know what kerning was. that confused me
Amazing hehe
Can Linus review the typography of the TVA in the disney+ tv show _Loki_ ? It makes use of typography really well and im interested in Linus’ take.
You’re an excellent pubic speaker, well done 🎉
Cool lecture. What’s the black serif font?
Migra Extrabold!
@@LinusBoman thanks a lot.
What are you using for those slides? Transition animations are super clean.
I'd love to know as well! I've been coding my slides from scratch to achieve a similar effect, but it takes a lot of work to get it like that.
I know this sounds semi-obvious, but Keynote actually has some great ‘smart’ transitions now that allow elements from different slides to be used in animations now or in combination with slide in/outs. Could be that.
Actually believe it or not - it's Canva! First time using it in years, and actually found the simple interface very helpful in putting together a clean and minimal presentation. partner.canva.com/timesnewboman
@@LinusBoman Makes sense! I've never given Canva a fair enough try thinking I could just do it all myself.
Thank you thank you thank you! After digesting hundreds of AI Art videos, this is the first true examination of real world comparisons that I have seen. I'm sure this is not unique (just like some AI Art) but the first I have seen showing the benefits of human over machine. I was so sure we had reached the promised land at the start of this explosion....but we haven't yet, have we. Ho hum, onwards and upwards.
I was mildly surprised when I saw Microsoft had added gpt3 into their search categories.
19:38 What OH GOD I DIDN'T SEE IT
I need "classy as f*ck" on a shirt 😂
I wish you did zoom in to those cursed generated images.
Fortunately or unfortunately depending on who you ask, in the US at least, such imagery generated by artificial intelligence can't be copyrighted, so now the real art lies in cooking up a more convincing image.
I love your stuff, can you help me please?! I need to whip up some political campaign signs, what fonts can you point me towards? love your channel, help me out in short term, and then maybe spin it into a video for later on. I'm US based so we have our political season upon us, and frankly I'll just head to the corner for some design inspo. hope you're well!
But you should definitely checkout Levende Streg channel to improve your prompting skills.
I find it hard to take AI generated art seriously, especially after the Lensa debate recently. Gathering data is one thing, gathering it from artists who don't consent their images and styles to AI system generation is another. It's a new technology, but it needs to be regulated. I can already see artists sueing AI prompt engineers for using their work without consent.
AI tech is becoming very powerful, but our current design and methods for AI will never be able to create, understand meaning, or truly become intelligent, no matter how big the neural networks become. I think this is what you meant by the AI "struggling with symbolic complexity". They can mash images together with better and better accuracy, but they'll not make deliberate creative decisions. This is not exclusive to AI for art, but all AI.
this isn't aging very well
Just ask it to draw a hand. 😂😂😂🤣😂🤣🤣🤣🤣🤣😂
I am annoyed by the dangling e's at 6:45
First