OpenAI DALL-E 2 - AI or Artist? Which is Better? 🧑🎨
Рет қаралды 174,278
Күн бұрын❤️ Check out Lambda here and sign up for their GPU Cloud: lambdalabs.com/papers
📝 The paper "Hierarchical Text-Conditional Image Generation with CLIP Latents" is available here:
openai.com/dall-e-2/
🕊️ Follow us for more results on Twitter!
/ twominutepapers
🧑🎨 Check out Felícia Zsolnai-Fehér's works:
/ feliciart_86
🧑🎨 Judit Somogyvári's works:
www.artstation.com/sheyenne
/ somogyvari.art
🎤 Beardyman (note: explicit):
• TOM SCOTT came to my s...
🍩Andrew Price, Master of Donuts:
• How I rendered 17,731 ...
😸Grumpy Cat ready for takeoff:
/ 1534560293069676546
🦊Fox repaint:
/ 1533724395281481729
❤️ Watch these videos in early access on our Patreon page or join us here on KZfaq:
- / twominutepapers
- / @twominutepapers
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Balfanz, Alex Haro, Andrew Melnychuk, Benji Rabhan, Bryan Learn, B Shang, Christian Ahlin, Eric Martel, Geronimo Moralez, Gordon Child, Ivo Galic, Jace O'Brien, Jack Lukic, Javier Bustamante, John Le, Jonas, Jonathan, Kenneth Davis, Klaus Busse, Lorin Atzberger, Lukas Biewald, Matthew Allen Fisher, Michael Albrecht, Michael Tedder, Nevin Spoljaric, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Rajarshi Nigam, Ramsey Elbasheer, Steef, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Timothy Sum Hon Mun, Torsten Reil, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: / twominutepapers
Thumbnail image was created with DALL-E 2.
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Chapters:
00:00 What is DALL-E 2?
00:54 Team 🧑🎨 vs Team 🤖 - Round 1
02:36 Team 🧑🎨 vs Team 🤖 - Round 2
03:37 Team 🧑🎨 vs Team 🤖 - Round 3
04:29 Team 🧑🎨 vs Team 🤖 - Round 4
05:14 Team 🧑🎨 vs Team 🤖 - Round 5
05:41 An important lesson
06:00 Grumpy Cat in space!
06:36 Fox Scientist
07:00 Platypus experiment
07:28 A visual pun
07:52 A time traveler
08:10 All hail the mighty Cyberfrog!
08:19 Rosie the Paper Dog
08:40 An image of you Fellow Scholars!
09:12 History in the making
Károly Zsolnai-Fehér's links:
Instagram: / twominutepapers
Twitter: / twominutepapers
Web: cg.tuwien.ac.at/~zsolnai/
#OpenAI #dalle
@generalfishcake 2 жыл бұрын
I'm a game artist and it is very exciting to have this as a tool. It will never replace really complicated and detailed pictures (for example, Grzegorz Rutkowski, Ruan Jia, Piotr Jablonski, Wenjun Lin, etc.) - but it can greatly speed up the sketching/blocking in process during the initial phases of work. Fantastic!
@solarpowered3364 2 жыл бұрын
Never say never. When people say that about AI it's almost certain it will happen
@Revvnar 2 жыл бұрын
That's wishful thinking my dude. AI is literally a programmable brain like computer. Within a few years at most it will go beyond even those artists you mention. It will easily be able to generate "Mandelbrot zoom" level like detailed art :/
@brabes76 2 жыл бұрын
Yes this would be. Great tool to speed up the design process. I wonder if this AI can spit out spritsheets as a result. Or your Characters from different angles or poses. That would be mindblowing
@beniciobonilla2271 2 жыл бұрын
I like your way of thinking. This tool isn't meant to be something that replaces artists - I don't think human art will ever be something that dies, we've been doing it for thousands of years and we're not going to stop because of AI. I'm not an artist, so take my opinion with a huge grain of salt, but I would rather try to use this program for the benefits it could bring that would help me embrace the inevitable change and give myself a leg up, rather than lull myself into a sense of doomerism.
@maxington26 2 жыл бұрын
"It will never".... careful there. Most of us thought that AI would never achieve the level of Dalle-2 or Imagen just months ago. Yet here we are, and both the training and the research are not stopping here.
@ThioJoe 2 жыл бұрын
I'm most excited about this simply for generating obscure and specific images. Lots of times in a video I'll spend a bunch of time looking for a stock photo of something like "person staring confused at a router" with no results. If I need high quality then I'll go with a stock photo. But if I just need a quick concept image as filler content, this is perfect.
@laden6675 2 жыл бұрын
Just curious, how much would you be willing to pay for a service like that?
@ThioJoe 2 жыл бұрын
@@laden6675 Over $100/month easily, even more depending on how often I end up using it
@BeCurieUs 2 жыл бұрын
@@ThioJoe Ya, you can pay over that buying images and still not get exactly what you want
@Djorgal 2 жыл бұрын
To create DnD characters that would be so great. Not only can you get the specific character you want, you'd be able to adapt his gear. My warrior changed his sword to an axe? Got a scar? A haircut? No an issue.
@SineEyed 2 жыл бұрын
@@ThioJoe if this AI hits the market and gets a similar pricing model to GPT-3, an image returned for _"person staring confused at a router"_ will probably not even cost a quarter. Do you use 400 stock photos in your videos every month?..
@user-vn9ld2ce1s 2 жыл бұрын
3:38 - after a closer look, the ostrich is not really an ostrich, it's a dinosaur, and the joke is that he's driving on "fossil fuel" (the spoiler), I don't think that any AI is capable of making such exquisite dad jokes...yet.
@hivebrain 2 жыл бұрын
I thought it was a giraffe.
@arbaguincci7591 2 жыл бұрын
actually, he somehow didnt just mistakenly see it as an ostrich one time, he literally put the prompt into dall-e and compared the two, then edited it into a video, and never noticed. Seriously is this guy legally blind?
@user-vn9ld2ce1s 2 жыл бұрын
@@arbaguincci7591 well, no need to be that harsh, everyone makes mistakes...
@simond.2479 2 жыл бұрын
I was also seeing a reptile in that image rather than a bird (altough, they look close in that case)
@eldritchlemon 2 жыл бұрын
i thought it was a snake
@TabletopRPGMusic 2 жыл бұрын
I’m interested in how Dall-e 2 handles maps. Could you try something like ‘dungeons and dragons battle map of a mushroom forest with treehouses’
@TwoMinutePapers 2 жыл бұрын
Coming right up! Really cool prompt, loved it. twitter.com/twominutepapers/status/1537098423152922624
@topaz-rn 2 жыл бұрын
@@TwoMinutePapers Holy mother of papers!
@OperationDarkside 2 жыл бұрын
Visit his Twitter. He fullfilled your request.
@mewwww17 2 жыл бұрын
Wow, that was so fast of a response, and such a good one too!
@elsenseidelaregla3417 2 жыл бұрын
@@TwoMinutePapers dall e 2 can create cards like yugioh or magic the gathering ?
@Zeekar 2 жыл бұрын
The way Dall-e 2 demonstrates a general understanding of a face, but isn't very picky on how it draws eyes is fascinating to me. That is pretty much the opposite of how a human would draw, since we're fixated on getting that particular detail right, regardless of style (It also gets other bodily details wrong, but so do we). This is a fascinating example of ai disalignment outside of ai safety. It clearly doesn't understand what features we value quite yet, even while being highly capable. Edit: they cover this in their website: "We also used advanced techniques to prevent photorealistic generations of real individuals’ faces, including those of public figures" So my takeaway now is that nothing that comes out of Dall-e 2 can be taken at face value. Pun intended.
@KevinUchihaOG 2 жыл бұрын
i wonder if it has something to do with open-AI having been conserned that the ai will be used maliciously. They have limited access and seems to have strict rules (the dude on the channel 'pl9nt' says that he might lose his access if he do real people, war, politics, etc). So maybe they have purposely limited the AI's ability to generate faces. I mean, it generate dog and cat faces perfectly and other AIs can generate perfect human faces.
@Djorgal 2 жыл бұрын
@@KevinUchihaOG The fact that no one can use it is very limiting for progress.
@septicsauce322 2 жыл бұрын
@@KevinUchihaOG I HATE that if it is true. All they are doing is hitting the breaks while plummeting off a cliff. Meanwhile, a few years from now. Someone (else) will make the same program, but better and public. (Without restrictions)*
@MeGaDwarf2008 2 жыл бұрын
@@septicsauce322 >Meanwhile, a few years from now That's exactly why they do it. Because by the time this type of software becomes widely available, OpenAI might be able to create a viable solution to mitigate unethical uses.
@septicsauce322 2 жыл бұрын
@@MeGaDwarf2008 Oh no, the program won't be open ai. It will be some rando in his basement with his super computer. That's what I was trying to say. Open ai can make their mitigation, but it will have no impact on Creator 3.0 TM. Open ai's only solution would be to claim that this new company stole their product.
@0oh_no 2 жыл бұрын
AI is becoming indistinguishable from straight up sci-fi
@himanshugarg357 2 жыл бұрын
Ai is brining imagination to reality
@sebastianjost 2 жыл бұрын
AI is science, that can make your fiction reality. (depending on what you want, it may still take 5-50 years though.)
@andrasbiro3007 2 жыл бұрын
And soon it may reach Clarketech level (sufficiently advanced as to be indistinguishable from magic).
@edzehoo 2 жыл бұрын
nevermind scifi. AI is becoming indistinguishable from magic !! As what Arthur C Clarke predicted
@danijel124 2 жыл бұрын
I wonder what will happen when AI will be able to design new technology blueprints...
@tiefensucht 2 жыл бұрын
Back in the days I thought that the holo deck in Star Trek couldn't work by saying simple commands like "create an bar from the 90s with some cool people in it", even if we had this holographic technology. Now we are here. In 10 years we might have an holo deck in VR that understands speech, can talk by itself and create complex 3d environments by these simple commands.
@ThorNjord 2 жыл бұрын
What a cool idea!
@gabrielandy9272 2 жыл бұрын
atleast learn something by this, never say something will never work, go 200 years back, and u would have issues explaining internet computers and everything people would not belive you we never know how the technology will evolve, we did not expect Artificial inteligence to skyrocket this fast in the last decade.
@delphicdescant 2 жыл бұрын
10 years is probably optimistic, just because of the vast difference in training data size, training complexity, and getting your corpus of training data in the first place. I suppose it would have to be trained on a zillion 3D interactive environments, which someone would have to build, or find and label in some meaningful way, or something. It's a million times easier to get images with descriptions attached, and humanity has been building a massive corpus of them for decades. But maybe there's a shortcut. I hope so.
@subspaceanomaly 2 жыл бұрын
Google Imagen can generate 3d from a couple of photos. If you get the a.i to generate 3d from its own images then you're nearly there and I would say that's already possible. I think a rudimentary voice to vr environment is possible now if they join the dots together. Rendered through unreal engine 5 and we'd already be getting quite close to Total Recall type stuff, with a.i language bots voicing characters you interact with. It would be a lot of work getting all that up and running and the processing for it would be rather large, but I reckon if a big funded team of experts worked to create that, most of the tech is already there and what isn't could learn fast to fill the gaps. I guess one reason it might not happen soon is that there isn't much point in doing it until you could roll it out for large numbers of paying customers and that probably is a few years or more in the future.
@AntiAtheismIsUnstoppable 2 жыл бұрын
The next step on the AI ladder is already in the making, *_the rights for machines to vote._* When this is implemented, and it will if the politicians and people in power get their way, then the next step again is, that machines should also become rulers, leaders and presidents. Because why when you have the right to vote, you do not have the right to also participate directly in the political system as part of that system? I will warn anyone who believes this is a good idea, that you have no understanding of what hell is, untill you experience such a world.
@markmarketing7365 2 жыл бұрын
Hahaha om. I've always heard your intro as "This is Károly Zsolnai-Fe _here_". Only now see that's the "here" is part of the name. :))) Thanks for the videos, I love it.
@karotto594 2 жыл бұрын
Ever since I watched with captions enabled I knew
@NikiWinProd 2 жыл бұрын
i noticed too haha
@brian9801 2 жыл бұрын
lmfao I thought that too
@manuelmathiasmagalhaes9776 2 жыл бұрын
hahaha me2
@linkin543210 2 жыл бұрын
Sounds like you’re a bad AI nlp model
@iseetheendisnear2416 2 жыл бұрын
An artist’s job is not necessarily the production of content, but the combination of separate parts into a meaningful whole. Using photos in paintings is a common practice. I see this as a time-saving tool I can use to fill up the canvas before I start actually shaping the statue from the marble
@bluebutton36 2 жыл бұрын
You should ask the AI to draw "two minute papers". I'm really interested in what will come out :)
@ENJERUNE 2 жыл бұрын
As an artist myself. I see so much potential for this tool. I don't think human artists will be replaced, but I think Dall e would be used as inicial concept art, and then human artists would develop more ideas around it, which would make the process more easy and cheap specially for small independent teams. And also as an individual artist, I been using the mini version whenever I have an artblock and I have an idea that I cannot visualise. And it made my work more easy and productive. I can also see this as a tool commissioners can use to show visually their ideas to an artist they want to commission, which will make artists work so much easier. I'm so excited to see what this tool brings to the world when it realeses to the public!!
@maysachan9921 Жыл бұрын
@@rest_in_pain same for u
@marek_tarnawski Жыл бұрын
@@rest_in_pain See. There is other side of the coin here. Maybe there could be less demand for artists to do work for client but at the same time independent creators can have more image processing power to start their own thing. You can now become client. Want to make your own game/ comic book/ movie? It's going to be cheaper to produce.
@heijd 2 жыл бұрын
I wonder how it would handle topological prompts, like "A twisted mobius strip" or "klein bottle" or something like that. It would need to properly understand what property it needs to maintain.
@michaelleue7594 2 жыл бұрын
Not really. Anything that has a strong correlation between what it means and how it's depicted online will look fine in DALL-E, and both Möbius strips and Klein bottles fit that description. If you try something like "a slice of Swiss cheese with four holes" though, it will fail utterly, both because DALL-E can't count and because Swiss cheese depictions are typically shown as wedges instead of slices.
@fedyx1544 2 жыл бұрын
morbius strip?
@MH_Binky 2 жыл бұрын
This is absolutely amazing, but, even if I'm not worried about the idea of AIs getting better than human artists, I am concerned about them becoming "good enough". Artists would obviously still be able to get private commission work from individuals, but companies having access to this kind of tool would absolutely use it rather than having to pay a person if the results were good enough. And I have multiple artist friends who are already struggling to get work as it is.
@TheGoukaruma 2 жыл бұрын
This is dangerous ideed. This AI can make 100 diffetent versions of a promt and some of them probably will suit the company better than that of a single artist.
@BenoHourglass 2 жыл бұрын
@@TheGoukaruma There might be a brief period of "AI touchup artists" that take an AI-generated image and fix the issues that the AI still can't fix. However, then that "AI touchup" could in itself be automated by training the original AI images and comparing them to the touchups, effectively teaching the AI to fix its own limitations.
@kettlefleet829 2 жыл бұрын
capitalism
@TheGoukaruma 2 жыл бұрын
@@BenoHourglass Well even a talentless hack like myself could use this AI and say " I want the same but with [some change]". Even people with no talent can judge if something looks good enough.
@yoursexualizedgrandparents6929 2 жыл бұрын
Bad day to be a digital artist, pretty soon this is all we'll need, and we all know how much companies love work without pay.
@Alex-ck5gf 2 жыл бұрын
the same can be said for programmers , mathematicians , engineers etc, we have so much data in all these fields to do similar concepts, in theory good bye everyone
@yoursexualizedgrandparents6929 2 жыл бұрын
@@Alex-ck5gf Remember when we had to pay people to use a calculator, I mean I don't, but it still happened.
@James-cb7nb 2 жыл бұрын
@@Alex-ck5gf the only way those jobs can be taken away is with general ai. This art has hardly any constraints so it's easier.
@art2stay178 Жыл бұрын
@@yoursexualizedgrandparents6929 Sorry bro, bad day to be a programmer, soon this is all we'll need for programmers, and we all know how much companies love work without pay. kzfaq.info/get/bejne/istlpbKFyKuYemg.html
@TheArtofCodeIsCool 2 жыл бұрын
Super talented artists might be able to beat Dalle2 if they have 3 days to make something, but the fact that the AI can get 95% of the way there in seconds, for free, combined with the fact that AI is improving at a staggering rate means that most graphic artists will be out of a job soon.
@dmarsub 2 жыл бұрын
Well even for those purposes most companies will still have to pay someone to sort through results pick the best ones and edit them as needed aka an artists (maybe earns less) just to add a few factirs to this complex situation: increasing the value of curation over creation that the internet facilitates already. This will change a lot for small creators/companies who couldn't afford artists at all. and will open up cheaper (per picture not necessarily per hour) jobs for artists in nieches that didn't employ artists at all or can now afford more ambitious projects. the job of artist will change and the skillfloor will decrease a lot. this could lead to a lot worse and cheaper art out there for a while. it's qlways difficult to predict the future of a complex system if we just include one new factor (picture A.I) without looking qt feedback effects and developing markets. but it will certainly change the market and the job.
@jaguar4118 2 жыл бұрын
Sadly I think this is probably true. I imagine people will still do art for self fulfilment- but the time is soon coming where the AI will be far more practical for creating good art pieces.
@2darki 2 жыл бұрын
Theres a bit more to art than just just having the image. You might require all the layers of the art piece, a certain resolution or transparency, etc. Most logos are done with vectors, which are a bit more complicated than pixels. Also we can pull up the fact that all those AIs are trained with 90% unlicensed Images and basically stolen art pieces, which could make them commercially unusable.
@kvickart 2 жыл бұрын
@@2darki There is more to art, but I think the thing to think about is the fact that this is a democratization of the ability to generate imagery given a fuzzy concept. In the same way nearly no one cares to hire a professional photographer when you can just use your phone, we will likely see the same with art. They still exist, but in much less demand except for very high budget productions. It's not about AI replacing imagination, its about the fact that everyone will now have a leveled up skillset in the form of a magic button to generate any random idea that comes to their head. Also what you're talking about are technical details, but as time goes on more and more neural nets will be built to tackle remaining domains and tasks to be automated.
@camrodam 2 жыл бұрын
@@kvickart i think the photography analogy is spot on. The 80/20 rule will apply here, as well as the speed & cost factor to outcompete the bulk of low / medium skilled artists from paid work. Only the most creative and skilled will be able to earn a living from their unique signature work.
@kevinbenitez42 2 жыл бұрын
This really genuinely is a revolution in technology at the scale of how the internet revolutionized history. For 95% of people the AI generated images with no human assistance are more than good enough
@kevinbenitez42 2 жыл бұрын
@@ilove-jesus this paper has so many more implications than just images. Large blocks of CREATIVE work are likely to be automated like: websites, Books, videos, CODE with literally little mental effort from people. We are about to be saturated with mind blowing content for the price of pretty much nothing. All humans have to do is nudge the ai in a direction. Imagine triple AAA quality video games implemented by a team of 3-5 people this is insane
@kevinbenitez42 2 жыл бұрын
@@ilove-jesus I think it’s quite obvious it’s going to be a problem and there should be regulation but this conversation has been going on forever and I’m happy people are working on it. I just want to offer the optimistic perspective for a change. Regulation is coming Don’t you worry
@morgantorium 2 жыл бұрын
@@ilove-jesus got to agree to be honest, nothing ever seems to change until it affects the people who make money from this.
@kevinbenitez42 2 жыл бұрын
@@morgantorium you mean everyone?, pretty sure that anyone will be able to train and deploy machine learning models on their own computer or across multiple computers(blockchain?). Pretty hard to regulate and restrict usage if you ask me
@kevinbenitez42 2 жыл бұрын
@@morgantorium you mean everyone?, pretty sure that anyone will be able to train and deploy machine learning models on their own computer or across multiple computers(blockchain?). Pretty hard to regulate and restrict usage if you ask me
@andreanto8685 2 жыл бұрын
Many years ago, the invention of photography significantly dropped the value of realistic art. Such feats became less appreciated by the general public. Then, stylized art (which cameras cannot replicate) became the norm. Once stylized art becomes 'democratized', the value will drop as well. Cleaning up AI generated images doesn't sound like it's going to be a well paying career option.
@tincoeani9529 2 жыл бұрын
Thank god for me that it can't do animation or 3D models viewable in 360° yet x)
@jaguar4118 2 жыл бұрын
Yup you gotta be creating the AI to be benefiting
@hombacom 2 жыл бұрын
As soon as you want to have control of the result you are back to visual knowledge and artistic tools. Imagine the pain to have same scene or objects in different angles without adding random artifacts or different details. This is another tool to use, not the only way to do things.
@DavidHatesNewYoutube 2 жыл бұрын
Yup, if you're a working artist who's not in a lead position, your career is in danger. Even professional photographers took a massive hit when everyone got access to decent phone cameras, so I don't really understand the argument that AI will benefit working artists as a whole.
@avastorneretal 2 жыл бұрын
@@tincoeani9529 others can, and there's no problem of giving output of a one AI as an input to another.
@Art3mis-Cane 2 жыл бұрын
I remember when I first started using Google and would search weirdly specific high resolution renders just to see what I would find. My youthful imagination is going crazy right now with completely fresh ideas
@devfromthefuture506 2 жыл бұрын
Instead of comparing art with variations you should make a amazing test. Pay some artists on Five r to create art based on specific prompts and after the paid art is made by humans , you create art with the same prompt with Dall-e 2 and compare and record a great video for us. It's can even generated a research paper
@Aaron628318 2 жыл бұрын
And then assess it blind. Actually, present it blind in the video also...
@gutzimmumdo4910 2 жыл бұрын
using same prompt as u wuld to an human it would be a missuse of the AI, the AI is not a person and understand its own lenguaje, so that will already be a biased towards humans sucsess kinda test.
@TheGoukaruma 2 жыл бұрын
No, it would be a fair test. Dalle 2 does promts pretty well and in the end you want to see if AI can replace humans. Reading and unterstanding is part of the job.
@germangarcia6118 2 жыл бұрын
I mean, the human art he used in this video was pretty bad, tbh
@bujin5455 2 жыл бұрын
6:39. I actually like the AI fox scientist better than the human rework! I love the different eyes for example! Really gives the fox a LOT of character. I also like the blurry hand better! The work just has a feeling about it, like the fox is moving through time. The human rework might be more "controlled" but it loses a lot of the emotion in the process.
@SineEyed 2 жыл бұрын
I've been generating AI art for a few months now, and I appreciate the things in the images created that don't exactly make much sense, just as much as I appreciate the parts that are coherent to my prompt. I really like the abstract stuff - maybe because I think even if I could create images or paintings like that on my own, I wouldn't think to do these really weird and wonderful things that the AI does. Some of it is so bizarre - and I love it..
@DisturbedNeo 2 жыл бұрын
Aside from the difference in detail, in the AI's version, the shape of the head is quite different, the collar of the coat is a bit wider, and I just noticed he's casually breaking his own ribs 😂 The human version is much cleaner, but as a result much less wild.
@elijahagustino3605 2 жыл бұрын
How ironic, that the human creation has less emotion in it.
@bujin5455 2 жыл бұрын
@@elijahagustino3605 Indeed. Probably tells you something fundamental about the underlying nature of emotion. (Though, probably nothing we didn't already know.)
@JamilKhan-hk1wl 2 жыл бұрын
Yup its like his eyes were damaged due to his experiments or he made enchancements to his eye
@NicholasRehm 2 жыл бұрын
Can you see if it is able to generate video game sprite sheets? An example prompt: “super mario sprite sheet where Mario is dressed in a triceratops costume” and then convert the image to an actual sprite sheet file format (probably done manually…for now), upload to an emulator, and see the results deployed in the real game
@unrealredist4124 2 жыл бұрын
Well. It is adaptable ai taking reference of litterally millions of photos, in other words you will get that result but it (might) not work for animation
@dannyarcher6370 2 жыл бұрын
3:48 - That's a sauropod of some sort, not an ostrich.
@Zadamanim 2 жыл бұрын
I noticed that too, lol. It says fossile fuel on the car in 2 different places too.
@omegadecisive 2 жыл бұрын
Scrolled far for this comment
@Pellbort 2 жыл бұрын
@@omegadecisive TOO far
@Pellbort 2 жыл бұрын
Yeah, that's a dinosaur. Driving a car that says "fossil fuel."
@darymetal 2 жыл бұрын
What's crazy is that it can produce pieces of work that are actually at what i would call "advanced artist level".
@Tacsmoker 2 жыл бұрын
as advanced and 1000th of the time... rendering most artists obsolete.... and now its the reign of imaginative people!!!
@MatBat__ 2 жыл бұрын
This is crazy... so glad you got to use it to show us. Amazing as Always!
@peskyhusky920 2 жыл бұрын
Can't wait to put my hands on this amazing creation, have some nice ideas that could come out with interesting results.
@thecrazymoon6578 2 жыл бұрын
This is so great! So much art in such little time
@koko969w 2 жыл бұрын
shoutout to all the blender artists who learned from Andrew
@TwoMinutePapers 2 жыл бұрын
Andrew is the best!
@testpurpose5845 Жыл бұрын
I only stopped learning at donought 😂
@thegeekclub8810 2 жыл бұрын
I think what scares me is that while the best of the best artists will probably be able to win out against AI for at least a while longer, but for the average artist? New artists? This AI wins out handily. And by the time those young artists have reached mastery, AI might already be further ahead.
@Djorgal 2 жыл бұрын
Yup, and many companies need art by the bucketload. Sure, you'd want the main character in your video game to be designed by a human expert. But side characters? Procedurally generated content could also become better and better.
@hombacom 2 жыл бұрын
Visual knowledge and practice are always valuable. Great that something looks good but I also think it count to pay attention to details, and you don’t get that for free using AI tools.
@DavidHatesNewYoutube 2 жыл бұрын
Feels bad for artists who are starting out.
@bimajuantara 2 жыл бұрын
And I'm pretty sure it's happening to other fields as well (writing, language, music, etc). The question is, can the average human catches up with these ai?.
@thegeekclub8810 2 жыл бұрын
@@bimajuantara Frankly, I don’t know if the average person can catch up. AI is advancing at a startling rate, while human progress takes years of study.
@qq587 2 жыл бұрын
people say ai can't replace artists, well they sure can...
@withlessAsbestos 2 жыл бұрын
They are not only coming for your job… They will disguise themselves with your skin.
@glf454 2 жыл бұрын
Who picks the pictures?
@someguy2819 2 жыл бұрын
@@glf454 We'll get an AI to decide that
@qq587 2 жыл бұрын
@@glf454 i am not an artist, but i am sure that in the future anyone can make images with this kind of ai better than the images today's artists make.
@shiethegal 2 жыл бұрын
Nah, they can't and never will.
@Tigerhawken 2 жыл бұрын
I think it's important to compare the time it takes an artist to digitally paint something vs. how long it takes Dall-e 2 to render an image. It's a great way to make rough drafts, edits or base images that only need some touch-ups. Time is money, as most professional artists know, and this would cut our workload into a fraction of what it currently is. Clients wouldn't even need changes or feedback, because they could just put any image into this AI and do it on their own. It really levels the playing field in terms of pro and amateur artists, as well as giving people more output for their time.
@ericoschmitt 2 жыл бұрын
But also lowers the value and demand of pro artists... This will make a lot of artists even poorer.
@Tigerhawken 2 жыл бұрын
@@ericoschmitt I know, I have a degree in art, and I don't even have a job in my career field. Nobody respects liberal arts anyways, how much poorer can I get?
@free4fire 2 жыл бұрын
Just wait one or two more papers down the line a.k.a. when the product itself will be available, most of the time there won't even be a need for any artist to touch anything up. You'll probably simply highlight or mask the section of the generated image that you want to change (or perhaps combine from the various generated results) and tell the program how you want it and it will do it for you, only rarely will there ever be any more need for an actual artist, so companies can then just remove everyone except the art director and maybe one assistant artist to do the same work that previously 50 artists did. I'm willing to bet that in 10 years at least 85% of all digital artist are out of a job. Yeah might be a "black pill" outlook but so far those have proven to be the realistic ones, I'm also saying bye bye to me dream job here only weeks after I've finally decided to get back into art. :/
@DonVigaDeFierro 2 жыл бұрын
@@ericoschmitt I mean, art has never been a viable career path for everybody.
@nicholasn.2883 2 жыл бұрын
@@free4fire Go openAI 🙃
@wowiexist9875 2 жыл бұрын
I'm so excited for DALL-E 2! As a writer, I love the idea of inserting excerpts from my stories into this AI and seeing it brought to life.
@ekkehard8 2 жыл бұрын
Would you specify a same style you like throughout? Writers probably have so many more prompts to feed Dall E 2 than most people. Would you mind estimating how many prompts you'd give it in relation to how much writing it comes from?
@wowiexist9875 2 жыл бұрын
@@ekkehard8 Mmmm, well as for the first question, I suppose specifying the genre of the piece would help keep the style consistent? For example: painted in a dark-fantasy style. Or: In a noir detective style. As for the second question, I'm not entirely sure I understand what you mean, but if you're asking how often I would feed it a prompt as I write, I suppose it would be with the description of a new location or character, so depending on the pace and intensity of the chapter that could be 2-15 prompts per chapter I would estimate.
@rameshdevasi6720 2 жыл бұрын
my prompt will be_ "a robot painting Salvador Dali portrait, while Dali was on the floor, dead"
@Revvnar 2 жыл бұрын
That's and amazing summary, haha
@dantefernandez2455 2 жыл бұрын
Whoa! Long time viewer here. It's so cool you have access to such an amazing tool! I couldn't think of any better hands to put it in! Personally, my partner and I are curious how Dall-E 2 fairs with prompts for pixel art. We are currently making a video game and have experimented with other similar AI tools and found that its interpretations of pixel art are quite lovecraftian, or Picasso-esque at best. Would love to see a follow up episode or minisode either way! I'll be holding firmly on to my papers. What a time to be alive!
@abdoolzaidtoorabally4284 2 жыл бұрын
I've been playing around with the Dall-E mini and I have to say my mind was absolutely blown by the results of the prompts for 1. flamboyant calm 2. subjugated hope
@Onoma314 2 жыл бұрын
I set up a challenge for people, a gallery with 12 images, and I told people 6 were created by a human, 6 by Dall-E 2 ,all they had to do was pick which was which. However, after they completed the challenge I informed them it was a lie and all 12 were Dall-E 2 art
@cpikuski2924 2 жыл бұрын
Mad lad
@gokaytaspnar1355 2 жыл бұрын
Cool
@kvickart 2 жыл бұрын
As someone who once wanted to be a painter professionally, I feel like I somewhat dodged a bullet by switching directions. I understand that it's not as good as an artist (yet), but realistically people who may have once approached an artist will now resort to neural nets for a lot of visual development. While maybe you subjectively like one handcrafted from an artist better, the gap in some of these examples is so narrow that the human generally loses IMO despite the quality, because it takes so long to handpaint an image. Not only that, but now for paid work we may only need the very best artists to stick around. The scary thing is also you may be able to train a NN on an experienced artist's portfolio and generate work that would be like what you would have wanted them to paint for you. This tech while super exciting is also somewhat terrifying at the same time because we're heading into true uncharted territory in computer science and how it will affect society as a whole. Years ago I believed that programming would be taken over by automation well before artistic tasks, but it's looking like exactly the opposite. I feel like this is because art does not need the cold hard precision of logical designs. It feels like only a potentially short matter of time before we start getting NNs that generate things that require the logical precision that will take over more and more tasks we rely on people for
@gabrielandy9272 2 жыл бұрын
check the video of the open AI coding, they can do simple code already remember i don't belive the tech will stay as it is, i give 10 years and pretty much i expect it to be even better.
@LarsRichterMedia 2 жыл бұрын
Since we don't know where OpenAI took those 400 million images from which they used to train their machine we can definitely argue that they potentially already used the work of countles artists without permission. James Gurney became a Twitter trend because many people wrote prompts ending with "in the style of James Gurney". In my mind a gigantic copyright infringement that people either do not understand, probably don't care about or maybe simply want to ignore because personal benefit overrules ethics with new tech since the dawn of humanity.
@zimbot_KWB 2 жыл бұрын
In the Dall-E Mini web site, I experimented a lot with various prompts, and the one that led to the most stunning results was "Cthulhu killing Leviathan in outer space". I would love to see what the full version churns out from that prompt.
@thallmeister 2 жыл бұрын
I wonder if it can do something a bit more in depth like, "A Bear painting a Lizard painting a mountain range with horses and a sunset behind them." It would be interesting to see how it interprets the nesting. Thank you for bringing this interesting paper to light.
@natasha6867 2 жыл бұрын
thank you for this amazing content. please make more dall-e 2 videos, the possibilities are endless! you could try describing existing works of art as best you can and see how well it matches it. you can see how abstract you can get before it stops giving accurate results
@jc3777 2 жыл бұрын
"What do you think?" "I think we're out of a job" "Don't you mean extinct? " -Jurassic Park-
@LisaFenix 2 жыл бұрын
I'm a 2D & 3D artist in games, and this whole breakthrough has been really exciting, but with a strangely existential feeling as well. It makes you question your future value as an artist. However, I can't help but always be excited for progress, and we are in the midst of a new industrial revolution! Artists will always adapt and embrace whatever life throws at us, and this is no exception!
@geort45 2 жыл бұрын
It'll become problematic in some levels, for the "small client" that wants fast and cheap results, maybe ideas for a toy, a painting for their living room, a logo, a character for a book, etc, the AI will yield a lot of results for free and the client surely will find one of his/her liking among them.
@NirvanaFan5000 2 жыл бұрын
@@geort45 : yeah, artists on Fivrr gonna be screwed =/
@darksunrise957 2 жыл бұрын
Yeah, I'm a 3D modeler, and recently I've been using the "Dall E 2 mini" site for concept art inspiration (not anywhere as good as the real Dall E 2, but it can still work for inspiration). It's been making me think about how we might soon have an AI that can be given a photo, and it recreates it as a 3D model. It's not really that big a leap in logic, considering just how well this AI recognizes content, shape, structure, and space, that it could translate over to a real 3D space. Then you could have something like Dall E 2 to make the concept hooked straight into the 3D AI to make a model of it, and bam, you type in a description of an object and you have a game-ready, modeled, textured, (possibly even rigged, looking back on other videos on this channel), asset. Sure, topology might not be great, but if you have a built in decimation module in the modeler, it should work fine for static meshes. Also, the AI could generate multiple different models like it does for the art, so you can pick your favorite.
@JH-pe3ro 2 жыл бұрын
Like with the advent of the Internet and the ease of getting instruction and references from it, it's another form of public good. AI's been getting really useful lately.
@thegeekclub8810 2 жыл бұрын
I hope you are right. My fear is that these AIs will be locked behind a paywall or something similar because right now only huge tech companies can afford to build them.
@Poney01234 2 жыл бұрын
Imagine spending 6 years in Art school, graduate in 2024, just to find yourself out of work because a computer can do the same work as you, a thousand times faster and for free ...
@cappypyramsaudpate5535 2 жыл бұрын
and then try getting another degree only for that to be AI automated within 2 years. im pretty sure the elites in our society want us to start going broke by the millions because nobody has addressed this issue on an economical standpoint
@NewPhilosopher 2 жыл бұрын
Someone that qualified will be able to find a niche.
@Mutual_Information 2 жыл бұрын
Holy mother of papers! I wish I had a catch phrases like this. My edu content could use it 😏
@disarmyouwitha Жыл бұрын
I signed up for access to DALL-E2 and received it recently as well~ I am VERY IMPRESSED at how well it does faces now compared to DALL-E1. Fantastic.
@mantovannni 2 жыл бұрын
I'm on a waiting list to try Dall-E 2. I've made about 160 images with Dall-E just yesterday, it is so impressive compared to any others I have seen.
@remelin75 2 жыл бұрын
Will be nice when we can see a new movie from any book we like as the AI computes the scenario in a different style every time we watch it depending on how we react to what we see.
@AntiAtheismIsUnstoppable 2 жыл бұрын
Or, when AI is being used in court of law as judges, what can possibly go wrong?
@Rctdcttecededtef 2 жыл бұрын
God help us
@TKs3DPrints 2 жыл бұрын
love it i am on the list and cant wait to get to try it. also want to try the google one.
@realastropulse 2 жыл бұрын
I love your final take, it's exactly why I got into AI in the first place. As an artist, I can give it a description of what I want, and get a beautiful first or second draft for my final image. That skips 3 or 4 steps and saves many hours of time!
@raeldon 2 жыл бұрын
hehe it wasn't an ostrich it was a dinosaur! i love these videos, keep it up!!!!
@EduardoRohdeEras 2 жыл бұрын
It's exciting and scarry to have this powerful tool. I can't stand wait to use it one day!
@GraveUypo 2 жыл бұрын
@@rdb223 not as good is an understatement. it feels like it's 6 years behind
@verstraetenandre 2 жыл бұрын
Machines don't live, don't suffer, don't laugh, are not anxious or curious, and this is why they can't share all these necessary human emotions and create true art.
@lonewolfzor 2 жыл бұрын
I have been waiting for this video, i knew it would come eventually, and im glad.
@PhilTruthborne 2 жыл бұрын
Given how subjective art is and that there are a whole lot of aspects to concider i would say that both AI and human work has their own places where they work better. Progressing AI technology is a very good thing despite some people fearing being replaced, but in reality the AI will be mostly suitable for specific concepts and mass production. You can ask the greatest artist in the world to make art for you but the true image in your head reflecting your own expression can only be made by you alone and the effort going into the work is a part of what makes the art good in my eyes.
@jaguar4118 2 жыл бұрын
that is true, but I suspect that eventually we will be able to use AI to do that more successfully than we could do it ourselves. Think of how Dall E 2 now can edit certain parts of an image- maybe we would make the sketch, then describe the parts of the image, and then the AI makes the real pieces and you get to choose from the variations. Or edit it again and clarify as needed.
@AntiAtheismIsUnstoppable 2 жыл бұрын
Except that the 99% of "scientists" who are atheists and, therefore believe that humans are nothing but machines, and, therefore conclude that machines are also humans, and, therefore *_machines should get the rights to vote._* The most scary horrifying idea of all times and it is being adopted by politicians and people in power too at an incredible speed. Imagine for a short while, and you do not even have to hold on to your papers here, exactly who will be in control of those AI machines. If we implement this idea, humanity will be dead for good.
@Enju-Aihara 2 жыл бұрын
@@jaguar4118 this is fucked
@josiah42 2 жыл бұрын
"Art is subjective" is an overblown statement. If the AI is capable of generating work that people will pay for and generally consider high quality then it has the potential to enhance or replace artistic jobs. Simple as that. The consumers of art don't necessarily think that deeply about objective versus subjective. Emotional responses are pretty universal. That's why pop music works.
@Enju-Aihara 2 жыл бұрын
@@josiah42 do you really want art with no value
@tschaboo19 2 жыл бұрын
You have an obligation to show us more of this. You are privileged to already have access to this. Thank you for sharing!
@MarkoKraguljac 2 жыл бұрын
Yes, we demand it! :3
@PhilAndyGraves 2 жыл бұрын
I love you can change around things. I hope soon we can see the vectors, transitions, and splines.
@Zebred2001 2 жыл бұрын
The question I have is this. Is it possible to select a specific AI generated image and input suggestions to tweak or fine-tune the picture more to taste?
@darksunrise957 2 жыл бұрын
Well, one of the options the AI has is to paint out a section and describe what should be there, so at most you'd have to save the first generated photo, then pass it back in and do the painting mode on that. There's no reason it couldn't be integrated like you described; I'm just not sure if they've done it yet. It's the same thing I was thinking.
@Zebred2001 2 жыл бұрын
@@darksunrise957 Thank you. I assumed It'd be something like that. I have no idea how you get ahold of this thing to play with it. I am very interested in these things ultimately being applied to VR experiences.
@gutzimmumdo4910 2 жыл бұрын
@@Zebred2001 yes, u can do this with dall e 2 already, u can inpaint and u can also select an image u like and generate more according to that image.
@ingoos 2 жыл бұрын
If AI can produce such images, given a description, which AI is capable of producing animation /video?
@epicthief 2 жыл бұрын
that's a big next step, hope we see something soon
@RichOrElse 2 жыл бұрын
So far we only have AI generated music videos
@SeanKula 2 жыл бұрын
@@epicthief two papers down the line.
@billMcLatentspace 2 жыл бұрын
There are already some notebooks since VQGAN&CLIP that can make a sequence using multiple prompts and keyframes for frame movement etc. Things like Pytti, Disco Diffusion etc. But yeah as for just giving a single prompt and getting videos, those aren't really out yet but projects like NUWA have teased it. Interesting stuff
@LSHFT_Eido 2 жыл бұрын
it was done right here lol 😆 kzfaq.info/get/bejne/Zsx0fbue1sqyXXU.html
@Shredonus 2 жыл бұрын
Im waiting on waitlist, and I can't wait to try it by myself :D
@seedmole 2 жыл бұрын
I've been pumping out album/track art for my solo music project with discodiffusion for a week or two now, so impressive what it's able to do.
@somerandomautisticguy2181 2 жыл бұрын
I feel like my painting skills is useless when an a.i. can make art in seconds while i paint for hours
@Hoytehablode 2 жыл бұрын
Man I work at the engineering department of oxford univeristy for various research department including robotics and AI and this is still incredible. I have askednfor access to it but I don't know if I will get it and I am dying to test this out
@GeoffJop 2 жыл бұрын
I would love a series that involves dall e in this channel, really interesting
@fangthewarrior 2 жыл бұрын
As someone who draws this sounds awesome! I'd love to put my own OCs through it to see what variations it can make! I'd also be interested to know if it could *interpret* my drawings as easily haha
@lerchiosi 2 жыл бұрын
i never expected visual artists to be the first people replaced by AI. That really surprised me.
@clumsy_en 2 жыл бұрын
Can we make it a regular thing and post few of these generated artworks by DALLE- 2 right after mentioning sponsors at the end of all future episodes? Kindly please.
@artistryunicorn928 2 жыл бұрын
This is so exciting!! I think this will fuel the world with so much inspiration for art
@curtheisler1200 2 жыл бұрын
I'm so happy you got access to DALL-E 2 sir! I know you put it to good use. :)
@heart022 2 жыл бұрын
What would happen if you were to ask DALLE-2 to make an image of itself? How would the AI view itself?
@flynntaggart7216 2 жыл бұрын
Dalle is actually a name of a robot in a movie so it will output the image of that (it will be orginal)
@royalblanket Жыл бұрын
The robot's name is Wall-E
@theskoomacat7849 2 жыл бұрын
This is great! Any chance of this becoming publicly available like Dall-E mini?
@MikMoensted 2 жыл бұрын
It already is. Check out his previous video on the subject. Link in the description there.
@Nightstick24 2 жыл бұрын
This is just mind melting. The fact that it can read into the prompts so much to infer detail that is meant but not explicitly said is incomprehensible. It’s ability to pick up a drawing and work with it in revisions is incredible. I can’t even begin to fathom how this will change the world when it’s more widely available. Storyboarding and blocking in rougher conceptual art will be revolutionized. The creative process will be taken to a new level, and I imagine artists could likely do vastly more work at a vastly increased pace if the AI could hand them a good drawing that they just need to make great. Outside of the image generating though the underlying AI is also fascinating and I think could change just as much or more. Truly, this is an incredible time to be alive; for all the issues of today the technology just gets my heart racing with excitement, like a child on Christmas morning!
@abramsonrl 2 жыл бұрын
Got so excited from the papers I needed that sponsor outro just to calm down.
@jimvonmoon 2 жыл бұрын
Their marketing slogan should be "OpenAI... now actually closed."
@KEI-um1kc 2 жыл бұрын
Seriously a shitty move by them
@radekmojzis9829 2 жыл бұрын
People seem to miss the part that is a BIG deal about this AI. It seems it cracked the problem of machine perception - It sees the world as objects and their relationships, not as pixels... I would not say the problem of machine perception is solved, but this seems to be the first actual crack at it - and a big one too.
@hombacom 2 жыл бұрын
Still it’s visual parts and relations that AI doesn’t know what it really is. It often draws a hand with 5 fingers but not always because it doesn’t know how to count.
@radekmojzis9829 2 жыл бұрын
@@hombacom ... that is EXACTLY my point... if you look at cartoons, do they have hadns anatomically correct? - NO, does your brain still interpret it as a hand? - YES the point is not "but can it spell the word 'coffee'?", but rather that if it sees a cup with black steaming beverage in it, it sees it as a cup of coffee with all its implications and relations, not as "a bunch of pixels in a pattern i was trained to label as coffee"...
@hombacom 2 жыл бұрын
@@radekmojzis9829 I get the advantage. Though it's better to simplificate a hand or so to something round than wrong details, we human are quite good to spot them. If it does it good to a certain level it would be interesting to regenerate sub part of image to see if it has more "power" to do it correct a second time.
@sm71485 2 жыл бұрын
CONGRATULATIONS ON GETTING ACCESS!
@MrLarsalexander 2 жыл бұрын
It's a Dinosaur in the race car :) A baby Stegosaurus. There is also a cool connection to dinosaurs on the F1 car. "Fossil fuel" And the "Shell" logo is a fossil 😅
@c_ffeestain2570 2 жыл бұрын
This is a similar complex to Github's Copilot AI for programming (backed by OpenAI's Codex). There were lots of programmers claiming that Copilot and Codex were going to take their jobs, but rather, I believe, as its name suggests, Copilot should be used alongside a human coder to improve the quality of their code or to automate the redundant tasks they're always doing.
@andreanto8685 2 жыл бұрын
You are not looking at the bigger picture. The art industry is filled by a variety of skill levels. many lower skill level artists are making a living out of lower end commissions. These clients would have no problem accepting whatever dall-e can churn out. This will basically wipe out careers in this category. I can also imagine the 12 year old me, passionate about art, making the best dragon drawing i can do. But then my peers would mock it as they claim to be able to do better by typing a prompt on a computer. How soul crushing it would be for me? I would've probably given up and never become the professional artist that I am today.
@ericoschmitt 2 жыл бұрын
For now...
@ericoschmitt 2 жыл бұрын
This technology will completely crush many artists careers.
@M.Evra91 2 жыл бұрын
No. You're not looking at the big picture. Dall-E will reduce time and costs. With Dall-E, it will be so much easier for ANYONE to bring their imagination to life. Maybe in 10 years, people will be able to bring their anime idea to life simply by typing words on a computer. I wish I was still a kid. I had SOOO many ideas that i thought were cool but didn't have the skillset to turn them into an actual graphic novel or animated them. If Dall-E was a thing back then, me and and friends would have had a blast with it!! What a time to be alieve!!!
@DavidHatesNewYoutube 2 жыл бұрын
Copilot is a meme and a half anyways. Give it a bit though and we may see something on a similar level as Dall-E 2.
@LogicBeach 2 жыл бұрын
The text that these AI's generate really remind me of text from dreams, like it is really inconsistent just like a dream!
@michaelleue7594 2 жыл бұрын
Interestingly, it isn't inconsistent. It's just foreign. They've recently done a study on this exact issue and found that the text found in images specifically correlates with the content of those images, or of the prompt used to create it, at least under certain circumstances. There are nonsense words that have actual meaning to the AI: for some reason, DALL-E recognizes birds really well and has a consistent name for them, independent of art style or image content, for example.
@KaizokuSencho 2 жыл бұрын
It turns out androids really dream of electric sheep...
@Sebi076 2 жыл бұрын
Now this is quite innovative technology! What a great invention and innovation!
@Barely-Average 2 жыл бұрын
Great video! I'd be curious to see what it comes up with when you ask it to draw itself, or something more abstract as 'the meaning of life'
@SamuelVella1995 2 жыл бұрын
I remember hearing an argument for AI years ago, which talked about how all the dangerous and menial jobs would be replaced by machines and AI. However, creative jobs would remain for humans, now free of manual labour. This just destroys that theory. AI will replace everything we do, because it will be smarter than us and faster than us in every way. Engineering, artwork, writing, healthcare... We're making ourselves redundant. What is the point?
@alteredcarbon3853 2 жыл бұрын
That's why some people think the only way to stay relevant in the future is to modify humans into transhumans. Look at the Neuralink project for example which have the long term goal to merge the human brain to the AI capabilities. It seems crazy to some and it might not work, but at least they're trying something. Meanwhile the politicians in the world don't give a fuck about it and ignore the problem because they only care for the present. When you try to explain the issue to the average person they think it's just science fiction. UBI is not a solution, it's a band-aid for a transition. But a transition to what? We need a real plan for the future civilization we want to make. Global warming made us realize we need to unite and plan ahead and it showed how bad we are at doing these things. The world will react when it's too late.
@cappypyramsaudpate5535 2 жыл бұрын
the icing ontop is that dangerous menial jobs will remain for much longer. All of those sweatshops, mining, foresting, police, factory, delivery, and waste clean up jobs have no chance of being automated soon because not everyone can afford million dollar automation processes. But all of the white collar employers can afford replacing 95% of their workers with an AI with a monthly subscription of just $5. This is a step in completely the wrong direction and i feel that its deliberately on purpose
@amish613 2 жыл бұрын
Exactly...menial task are given to underpaid and severely over explouted workers who are often neglected and abused by lack of good strong labour laws in many 3rd world countries. Well, i find the future to be dystopian society. These countries don't have enough money to pay for govt jobs..only private sector left for jobs. If that also stops giving jobs...how can thr poor uplift themseleves from poverty? The competition increasing on daily basis....
@johnyewtube2286 Жыл бұрын
The way ai works is you feed it a large data set to train it. The researchers probably crawled Instagram or some other site for drawings and is using that data. Without that data the AI would not know anything. If people stop making art then the AI will not have new data to learn. This seems overhyped.
@BossKillRatio 2 жыл бұрын
Imagine a day when we can generate an entire video game with a text description.. idk it probably already happened for like indie side scrollers.. but its gonna be so awesome . Everyone will be able to play their dream games using AI as their slaves.
@hombacom 2 жыл бұрын
It’s hard to generate code for a game if you don’t know what to achieve in the end, AI has no knowledge what it feels to play a game.
@nemurenainda8848 2 жыл бұрын
@@hombacom That's what used to be said about art, and yet here we are
@hombacom 2 жыл бұрын
@@nemurenainda8848 Right or wrong pixels are not critical but wrong code would give bugs and errors.
@Fermion. Жыл бұрын
@@hombacom AI is already at an average programmer's level in abstract programming competitions. It's just a matter of time until it is able to write code to solve real-world problems, and write game code in real-time as you're playing the game. At that point, it'll likely create a much more efficient and dynamic programming language for itself, that we wouldn't even understand, which would minimize bugs and errors.
@hombacom Жыл бұрын
@@Fermion. you dont seem to understand, machine code is the optimized language for computers that we dont read but high level programming is for humans. AI solves very specific limited problems and has nothing to do with a end user program. Ai generates errors so no matter tool (like GitHub copilot) it’s the developer that is responsible for it. Describing a problem in natural language is never accurate and can be misunderstood (like dalle)
@i0am0superBlast 2 жыл бұрын
Man I want to get my hands on the full version sooooo badly. There's so much I want to see it try to make. I also find anime style artwork to be very appealing and Dall-E mini clearly knows what anime is supposed to look like, and even knows specific shows too. There are AI's that can make some impressive (although still noticeably odd) anime faces/heads, but Dall-E and the other similiar AI has waaaaay more potential. This AI is just so exciting that I can't even properly express it!
@Enju-Aihara 2 жыл бұрын
in the nature of most anime picture drawings, it will replace any artist forming it
@NoThrottle 2 жыл бұрын
I'm a digital artist and I am already using an implementation of DALL-E Mini for inspiration and references. This is a game changer
@sharkysharkerson 2 жыл бұрын
The results are impressive. I can see how artists can work together with the tool to brainstorm, etc. But I can also see how companies who are less discerning would rather test out a few prompts and get a generic image that fulfills their purpose, rather than paying true artists to make drawings from scratch.
@toms2oo8 2 жыл бұрын
At which point they're paying the developer of the AI/Service they are using. Honestly there will always be a market for "human made" art. The truth is though AI is eventually going to replace a lot of what we do. We just have to adapt and either work less (and get paid the same) or find new things to do.
@LovelyLori193 2 жыл бұрын
Curious how much energy is expended to create these images. Or the thousands of Dalle mini pictures I'm seeing on twitter
@MeGaDwarf2008 2 жыл бұрын
Not much really. All the real energy was expended when they were training the model. And that was a LOT of energy. After you have a finished AI model, generating new outputs can be done on basically any hardware with enough memory to store that model.
@iruns1246 2 жыл бұрын
@@MeGaDwarf2008 Isn't the model itself is huge though? If I'm not mistaken it has billions of variables.
@superfeel1275 2 жыл бұрын
@@iruns1246 still could run on most middle to high end computers, i tried running pre-trained models of various AIs and it's really not that expensive in terms of energy consumption, about what a graphics demanding game would need
@MeGaDwarf2008 2 жыл бұрын
@@iruns1246 Reading up on GPT-3 model size, I couldn't find an exact number, but estimates run at 400-ish GB of RAM. So you'll need a beefy server to run, but still accessible enough for a small crew or a very interested individual.
@jacky_west 2 жыл бұрын
Waiting for my access to dalle 2. I can’t wait !
@123arskas 2 жыл бұрын
AI was trying to tell in the text " I'm going to surpass all humans, What a time to be ALIVE "
@fabianmosele2321 2 жыл бұрын
Thanks for the kind words! I believe you're right, it shouldn't be human vs AI, but rather what humans can do more if co-creating with AI. Because at the end, the most amazing part of these tools is seeing what we can do with it :D
@Kabitu1 2 жыл бұрын
That's amazing... when is this sort of technology slated to become commercially available?
@SineEyed 2 жыл бұрын
OpenAI has a comprehensive website. You could learn a lot if you were to visit..
@LutraLovegood 2 жыл бұрын
Right now. There are already a few websites with this sort of AI, though none of the ones I tried were as good as Dall E 2.
@Kabitu1 2 жыл бұрын
@@SineEyed All they have is a waiting list to access their servers, nothing about plans for plain commercial licensing of their weights and biases.
@thesoundsmith 2 жыл бұрын
The Dall-e mini has kept me occupied for hours, what I would LOVE is the ability to select one of the nine images and say "drill down into this one," like you are doing. Mini seems to have a "New Yorker" default style, which is quite artsy, IMHO.
@RogueAI 2 жыл бұрын
DALL-E 2 can do that. You can select one of the pictures you like and hit "variables" to make a batch of six more in a similar style. You can also highlight a detail you want changed and tell it what to do.
@SethOtaku 2 жыл бұрын
What a time to be alive! I can't wait to have my hands on it! Do you know when one could get access to it? I need it right now!
@woowoo111111 2 жыл бұрын
Correct me if I am wrong, but isn't Judit's picture depicting a dinosaur driving an f1 car, not an ostrich?
@hmmmm7468 2 жыл бұрын
Imagine the future where people take credit an sell Dall-E pictures for absurd amount of money Or using Dall-E to enter an art school or something
@LightVelox 2 жыл бұрын
I mean, if someone is willing to pay, and even if they get into art school it would just be problems for them when they have to draw something without assistance anyway
@poopoodemon7928 2 жыл бұрын
People talk about how artistic and intellectual jobs are safe from AI. Rip to that idea. We need to be ready to restructure how our society fundamentally works, this is bigger then any of the past industrial revolutions.
@Enju-Aihara 2 жыл бұрын
@@poopoodemon7928 as an artist, should i stop drawing pictures?
@poopoodemon7928 2 жыл бұрын
@@Enju-Aihara I mean if you're only an artist because you wanna make money then the soon to be market might become more competitive. But if art is a hobby of yours then not much is changing. In other words, you now need to take into consideration that finding work as an artist might get even harder in the future. It's really up to you and I can't perfectly predict the future, but I hope I helped in some way.
@Enju-Aihara 2 жыл бұрын
@@poopoodemon7928 no, not for the money, but at some point i'll go offline because i can't differentiate human made stuff against computer made stuff. Those pictures i have and love now, is everything human made. Even filming what you are doing, at some point the ai can generate that, so i guess i will download everything before it gets lost forever.
@GraveUypo 2 жыл бұрын
I used AI to help me compose a track for my game. I have it the intro i came up with and it spewed out a absolute banger of a chorus. i just had to bridge the two parts from there and i love how it turned out.
@rafaelestevam 2 жыл бұрын
"What is my favorite, then?" I could swear that an incredible caustics was coming 😁
@Patschenkino 2 жыл бұрын
3:40 That’s a snake.
@loljptrollergami7325 2 жыл бұрын
As an artist this really scares me lol, I hope I don't get replaced by an ai in the future
@Aldraz 2 жыл бұрын
Creating AI is also the art, probably the highest form of an art. So if you think about it like that then artists may survive. Otherwise, yeah you are pretty fucked, sorry mate. But take it as a positive thing.. it may replace art now, but soon it will probably replace every other profession anyway. And that's gonna be either the best or the worst thing for the humanity. Let's be positive mate!
@Purplesquigglystripe 2 жыл бұрын
No you won’t. This model uses preexisting images taken from the internet to learn what stuff looks like. OpenAI does not have the rights to any of these images. The issue with that is there is no transparency on how close these images are to the preexisting images, so the model could be ripping off other artists or even committing copyright infringement. A model that could be used commercially would be way more limited in what images it can be given to learn from. Perhaps in the future, ai will be restricted in what images it can use due to copyright. Also the images it makes aren’t perfect and will still need actual artists to curate and finish.
@Aldraz 2 жыл бұрын
@@Purplesquigglystripe That's not how it works. Sorry mate. But these images it needs to train on are picked carefully to not have any watermarks, etc. They need to be correctly labeled and are often in already pre-made packages of images that people can freely use and download for these large AI models. They are also distorted or processed by many post-processing techniques. Then they use CLIP or latent diffusion and boom, you got DALL-E.. but honestly Imagen from Google made it much simpler and with better results anyway. And these talks about copyright are so funny, because AI is creating new images and not using any "filters" by any means. And even if they did, it's actually legal to modify copyrighted image, if it's so much different to the image before. Also, AI will soon generate almost all possible artworks humanity can come up with and therefore copyrights will lose it's meaning as there'll be absolute abundance of art and graphic designs. That means copyright will maybe not exist in like 5 years.
@Shadowed007 2 жыл бұрын
@@Aldraz well said.
@alansmithee419 2 жыл бұрын
7:45 There is a way you could interpret this as a visual pun (bare with me): Beneath this metal... thing... appears to be a pile of dirt/soil/dust. One possible meaning of "pun" is to consolidate earth by pounding it. And so you asked for a visual pun, it gave you an image of something being punned. I imagine the text, if it knew how to make English words, would be a joke, and so it is being punned into a pun, making the whole image a visual pun - by exploiting two possible meanings of the word "pun." To call this a stretch would be an understatement, I of course realise. "Spaghettifying" may be more apt. But I just love the fact that it does, technically, work.
@SocksWithSandals 2 жыл бұрын
I've been addicted to DALL-E since the previous episode.
@MyAwfulUsername 2 жыл бұрын
Bro that is a dinosaur not an ostrich
Kasim Tech
Рет қаралды 13 МЛН