Thanks for watching! The video above is a primer on how we got here, how this technology works, and some of the implications. And for an extended discussion about what this means for human artists, designers, and illustrators, check out this bonus video: kzfaq.info/get/bejne/qaxymdWLkrSXd2g.html

This is so scary, it’s only a matter of time before AI can generate videos and then whole movies with complete soundtracks, characters and a plot

As a researcher working on generative models, this is one of the best, clean and concise explanations for the tech! Kudos to the Vox team! :')

Seeing AI unfold in real time over the years is so satisfying and also a bit terrifying.

I feel discouraged from studying art. There's a feeling of satisfaction that comes from conquering something so difficult, from sitting back and seeing the fruits of my efforts grow. I'm afraid of becoming obsolete in a way that's difficult to articulate.

This is the best explanation of the tech I’ve seen so far. Would love to see a follow up video on this for animations. I believe this is a game changer for music videos.

As a digital artist and graphic designer, this is ridiculously fascinating and scary🤯 I've been watching every videos of Dall-E on the internet...

Reminds me of that one scene in I, Robot: Will Smith: "Can a robot paint a masterpiece?" Robot: "Can _you_ ?" At this point it can now be rewritten as: Will Smith: "Can a robot paint a masterpiece?" Robot: " *Yes.* Can _you_ ?"

As an ML researcher, this was the best 'public facing' explanation of latent space I've ever seen. Good job Vox team.

Joss is amazing. This is truly the pinnacle of tech journalism.

I used to daydream about computers being able to create a virtual reality version of a book. Looks like that’s on the way. It will probably be even better than that, with the simulation being customized for each user’s specific interests. The future is nuts.

I'm doing a research project in machine learning, and I've seen various KZfaqrs getting things wrong in their explanation of AI. But you guys completely nailed the concept of latent space. I guess you left out how exactly the encoder and decoder works, but this video is targeted at the general public, so fair enough.

The quality with which Vox is able to simplify hard things is amazing

you can’t just hit me with the fact that 2015 was 7 years ago so early in the morning

As an artist this scares and impresses me. Its scaring me cause of even the artistic field being overrun by ai in the future but impresses me cause it is just, impressive

I remember a teacher from my childhood once she said to me "your drawings will get you knowhere, they will save you from nothing", while embarassing me in front of everyone and ripping appart my drawing, and now, having battled heavy depression for many years and trying my best for working in the future in something that has to do with drawings or illustrations... AI came in, "learning" so extremely fast that in a handful of years they will totally outclass artists I mean, ofc I can draw just as a hobby, but a hobby would not give me enough money to pay my bills if the ones who would pay me look for a faster and cheaper option, I guess my teacher (and my father) were right all along

Can’t wait until this tech is applied to video game development. Imagine hopping online with your buddies on some FPS game, typing in a prompt about what kind of map you want to play, then having a completely unique and interactive map to play on.

I guess, once this model is perfected, the definition of a human artist changes to something completely different from what it used to mean.

Let us still not forget the fact that the generated images are derived from hundreds or thousands of creative artworks by us, humans. It is like the arts of different artist all come together to form an unimaginable piece we do not expect. It is still incomparable to an art piece that has the soul and passion of an artist.

Challenge, having done a Computer Ethics and Law Course were we covered some of the topic, is as follows When looking at sites like Facebook, Twitter, DeviantArt, or any other website that will host art content is that the individual artist or creator is not paying for the server space. Server space being what the buzz word is of "Cloud" which is owned by another company or individual. As such, the issue being that Meta (Formerly Facebook), and others own what ever a person posts to the social media platforms as it is on the company servers...that does not mean the creator does not have copyright to the creation, but that is the sticky situation when artists post works online period and can be found with a simple google search...the most similar form to understand is owning a physical copy of a movie, the person does not really own the movie and is not legally allowed to alter the movie from what the creator intended (though this too is being fought in the courts with companies like VidAngel trying to make even rated R films family friendly through editing like a non-movie channel can through a cable or satellite tv service can...through contracts that allow such actions). As the professor instructed, in the Computer Ethics and Law Class, to help protect images from being used for AI or even just being used without consent of the original creator...use Watermarks all over the image, as the AI and Machine Learning will not be able to get passed all of the watermarks to generate the clear images...also as stated in the class, instead of posting to Social Media on server space you are not in control of (as Facebook and them own it) everyone has three viable options with the last one being the worst of the three to show art works or any other creations that are the best protections from a simple google search that is helping to feed the AI generation of projects. Option 1: Host a personal server with your own personal images, and pay for a domain to show images you want, and due to SEO (Search Engine Optimization) understand you still will need to use watermarks on the posted images and then have a link or email where people can contact to ask for proper permission to use the art created without watermarks, but make sure to get a signed agreement that the person will not use it for personal monetary gain (similar to an NDA) Option 2: As most people can not get the licensing to host a personal server that is publicly available go through a website hosting company to host your personal website (NOT for FREE), pay for a URL, and follow the rest of option 1 Option 3: Use Social Media like Facebook, DeviantART, Twitter, etc...but make sure you keep original copies of the work...and only post once a large portion of the work has water markings to make it much harder for the work to be used by AI and ML when it comes to AI art.

When thinking of professions that could be phased out by AI, artists never came across my mind. The future really is scary.

As musician I'm excited and afraid in a same time knowing this will certainly come too for my art. What a time to be alive!

The question regarding the copyright of images generated by trained models is absolutely resolved! The precedent comes from a case called Naruto v. Slater which ruled that only a human can create art that is copyrightable. No one can own the rights to an image generated by an AI because they are uncopyrightable.

As a visual artist, thinking about the infinite possibilities for reference choices is very fascinating to me. But as a black woman who creates majority of my art around this subject, I am slightly worried about some biases in AI and how useful some of the generated images will be for me.

We humans derive meaning, satisfaction, hope and even therapy from conjuring up our own images. There is so much mystery within the frontier of our own minds but ultimately, it is limited. We do not want to find the limitations of our own psyches by watching AI outstrip and outperform what should be a human frontier. AI acceleration is dwarfing its own creations - never mind how small, mundane, and slow it could render human art. I’m a big fan of technology, but I’m a bigger fan of art.

This video is mindblowing and awe inspiring. Such simple and effective descriptions of the processes, along with original art created by the models!

As an artist who dabbles in the Arts as a photographer who dabbles in photography for a decade as a filmmaker who has dabbled in filmmaking for over a decade I am very excited for this new technology I think we should always embrace the future

As a science fiction fan, I always assumed that the time when AI technology advanced to the point of artistic expression would be after my lifetime. Yet, here it is in a rapid breakthrough when I'm only 60.

Incredible how this is progressing it's shocking what the researchers have done.

You've done a really great job here. The context & framing, the explanations, and your overall take, focused on the artistic outcomes that have already been happening in the short time since the release of these models, all A+. Motivating multiple dimensional representations with your balloon/banana classifier, and explaining latent space as a natural outcome of making progress on that task, I will definitely use that. And tying it all together with the cheeky "multi-dimensional latent space" Midjourney prompt 👏

I think this was bound to happen. I tried using one of these programs and it's actually cool. I think that having ways of giving credit to artists that are used for these programs would be a good step. I can see AI art being a great source of inspiration and ideas to use and plan out different types of art. For a novice like me, it can actually be a good learning tool over how to produce different types of art. This won't be an easy transition, but I think it's a necessary one that we need to hammer out the legal and ethical details now rather than waiting for them to haphazardly come together.

The quality of VOX's non-political content is BONKERS. So amazing.

6:38-9:49 That was by far the best explanation of deep learning I've ever heard. Also it gives you a sense of why mathematicians care about higher dimensions; their results are useful even if these higher dimensions don't physically exist.

this is another way of preventing creative blocks. imagine you're stuck in a scene you're creating for a chapter for a fiction book you're writing, and you're stuck with only a bunch of words to describe it. putting those words as prompts in a dall-e based image production system can help even writers visualize the scene better and even come up with distinct personalities for the characters they's like to have for that particular scene. and that's just one way of using it in a manner that could help expand creative horizons. film makers and musicians can benefit from this too.

One of the best Vox reportings I've ever seen. Not even Tech KZfaqrs are bringing up the potential consequences of this technology and you guys touched on it beautifully. Well done, very excited and scared for the future of this technology!

This just made me realize again how much more terrifying technology keeps getting

As an artist this is just incredibly scary.

I can't remember the last time I was truly mind-blown like this.. Really feels we're at the frontier for a new approach to creative arts, the ramifications are awe inspiring and a little terrifying. Also poses the question; how does this change how we feel about existing art, when you can create new masterpieces in an instant..

"The field of AI is highly interdisciplinary & evolutionary. The more AI penetrates our life and environment, the more comprehensive the points we have to consider and adapt. Technological developments are far ahead of ethical & philosophical interpretations. This fact is disturbing. We need to close this gap as soon as possible." ~ Murat Durmus (THE AI THOUGHT BOOK)

As a Computer Science (AI) student from the University of Toronto (where the paper in 00:26 came from), I'm truly amazed by how well this video is structured and how you intuitively explain the concepts of features, data, dimensions, latent space with such clarity! 👍 Thank you @Vox @Joss Fong!

So cool, so scary, so real. The biases, the creative possibilities, the 500 variables humans can even name. The fact that we can relate to the images and have such a clear process, can help us learn about our own creative process.

12:57 "my 5 year old drew war" is something.

I think this video explains most fundamental steps of the learning and creation, not only one specific part like 'the diffusion model' like in many other videos. Great work.

I wish you guys showcased DALLE 2 more, it's leaps ahead of Midjourney, although what they've been able to accomplish so quickly is amazing.

The last couple of days I have been creating images with chatgpt4+dalle3 and watching this video now I realize that even though it's only one year ago. Dalle 2 was the wright brothers airplane. While chatgpt4+dalle3 is a space shuttle. That is how astonishingly fast the evolution is going. The S curve is going bananas. How long will we improve at exponential speeds? Nobody knows. What will the cutting edge AI technology be when the S curves starts slowing? Nobody knows. For now, enjoy the ride before all the negative aspects of this will hit you like a crippling anxiety.

I feel bad for human creators that don't want their work used by the algorithm. There's no way to prove that it was their creation(s) that was used to make the image. The problem is compounded when you add in that every image generated is a true "1 of 1. Still the technology is truly fascinating.

Thank you for creating this video in such objective tone. I think this is a great starting point for anyone who wants to understand what's going on, whether they're in favor or not towards the use of this technology.

I just realized something really disturbing. If AI can match an image to a word, how are those "prove you're not a robot" tests going to work now? If AI can figure out what a tree is, "click all the tiles with trees" no longer works.

Are we lose the meaning behind the work? The meaning of the process of making art.

Best simplified explanation of deep learning I've ever seen... well done!

Coming from a computer engineer: you all explained this so well. This is amazing!

"i asked AI to make a Music Video... the results are trippy"

Just watched this today while trying to learn a way to explain this tech to non technical people I know and I am really impressed. Excellent work with a great breakdown with some awesome visuals.

As someone working with ML, I've gotta pay my respects to this video. Clearly explained and beautifully produced.

My simple mind cannot even comprehend how cool it is. Thank you Vox for explaining it so simply😘 Maybe in future, you can write and create a whole movie

1:23 It's been quite dramatic how the community of AI generated imagery has evolved since the release of this video

i gotta appreciate how this video hits the sweet spot of being informative but also simple. it's not so complex that it's hard to understand, but it's not so simple that it becomes misinformation either

It's exciting yes, but there is already a huge problem with plagiarism in Art. With the rise of NFT's many have taken artists work or even made an AI data set of the artists style to sell online.

As a digital collage maker, this is great for us as a never ending resource for material to use

OMG the best part is the reveal of Joss ! No AI can come up with such a beautiful image

Midjourney is still blowing my mind on a daily basis. And it's improved massively over the last month, the top images are incredible, way better than they were a month ago. And it's ability to replicate artist's style is amazing. This thing is going to put concept artists out of work.

one of the best videos I've come across on KZfaq, thank you to all involved, AI is progressing alot faster than I expected or times moving fast hehe.. amazing 👏

about 10 years ago, I was talking to a friend of mine and I was saying that if there is any technology that could turn thoughts into arts, designs, and music, I would become a great artist. But I had this thought more than 10 years ago! and now they are about to bring this tech to the real world!? That's so cool! Hats off buddies!

This channel is always so well thought-out and explained, it’s awesome.

Absolutely amazing video, from the investigation to illustration and the clarity of ideas, everything is perfect from this. I will say both congratulations and thanks for such content.

From a defendant's point of view, one argument could be that the use of a dataset of images to train a text-to-image model does not constitute copyright infringement because it is a transformative use of the original images. The purpose of the text-to-image model is not to copy or replicate the original images, but rather to learn from them in order to generate new and unique images. Additionally, the text-to-image model does not create exact copies of the original images, but rather creates new images that are influenced by the style of the original images but are not identical to them. Another argument could be that the use of a dataset of images to train a machine learning model falls under the fair use doctrine of copyright law. Fair use allows for the use of copyrighted material without permission for certain purposes, such as for the purpose of commentary, criticism, news reporting, teaching, scholarship, or research. As for the copyright laws in a class-action suit, the plaintiffs would have to prove that the defendant's use of their images in the dataset that was used to train the text-to-image model was not fair use. They would also have to prove that the text-to-image model created images that were substantially similar to their original images, and that this similarity caused them significant economic harm. It's also important to note that copyright laws differ from country to country and it's important to consult with a legal expert in your specific jurisdiction.

The explanation for latent space and deep learning was really well made imo!

This is an amazing resource for artists. You can generate an image to use as a reference for an abstract idea you have, and no one will ever get the same image due to randomness in the code.

James Gurney’s point of the artist being able to opt out of having their art as an input is totally fair. But I think this is really beautiful and cool. I also think it’s really good. I think giving everyone the ability to realize the contents of their imagination is really positive.

I dont think you intended to make latent space sound scary, but it definitely is an amazing thing to grasp on a philosophical level.

In terms of how generated images would affect artists, I feel like it might be similar to industrialization and mass production. There will always be a demand for real people's work. It won't replace artists just because it can generate anything you can think of. It can't think of its own, it just takes what has already been done and combines it. It doesn't know how to make a new art style, but real artists can.

The tune at 1:20 sounds like a remix of Toccata and Fugue in D-minor.

Fascinating... and terrifying and the same time. It also reminds me of a movie from 1956 -- The forbidden planet -- where a machine was capable of creating everything the inhabitants of such planet could imagine.

Yeah, it's a bit of a problem that the images of artists are used without consent by for-profit companies... Aside from that cool technology, it's gonna replace a lot of art but, at first, most probably, only lower-grade art. The interesting thing is that even if an artist's art is of higher quality, it will be overshadowed by the millions of images AI can generate specifically for a consumer in a blink of an eye. So, our art will become irrelevant for most viewers. But still, it's exciting to see how it develops!

I kinda feel bad for artists and photographers who tried their best to make the same results

Another excellent piece by Vox. And so much has changed in the last eight months, with AI, since this video was posted. Thank you for explaining to me how the back end works. Talk about complexity. And yes, like one of the gentleman‘s concluding words stated, “It will bring a lot of good and a lot of bad.“ Just like the Internet. But I think at a whole new level. Example: AI is generating photorealistic people (soon to come videos). Imagine AI generated deep fakes. Add to that the epidemic we have now of sophisticated hacking. Add it all to a pot, stir together and you have the potential for unprecedented social chaos.

give anyone who is responsible with this explanation a raise, you successfully explaining nth dimension that is easy to digest!

This is an amazing piece of journalism! Thanks for a succinct and well-laid-out explainer on the latest in ML art tech.

This is really well researched and love how many Interviews y’all did!

I'm currently studying digital art and I'm scared for my life.

Watching this video 10 months after it was released feels like watching a documentary from tens of years ago.

The level of anxiety for tech vs art is not new, a similar existential crisis took the hearts of some painters with photography. I'll not go to say it's the same, the speed of the revolution is so fast that will be more traumatic but at the end we'll find new ways of sharing our obsessions

As an artist I’m a little scared of this technology but also as someone who can appreciate technology this is so cool and I can’t wait to see where it goes from here.

I'm so envious of my nephew, he'd be my age when these have matured

You did such a great job. What a great overview. I particularly love your explanation of latent space. Very economical. Very clear.

Thanks Vox, loved learning the science behind it! For dalle 2, i believe mkbhd did a video on it

Really good video. One small thing: diffusion happens in latent space. Mapping to pixels happens at a later stage, which in Stable Diffusion, for example, is in the variational autoencoder (VAE). But this is as close as I've seen a pop-culture video get to describing latent space, so that's just a small nit.

Amazing. Thought of this before as I tinker IBM Image recognition, but they made it possible. The artists were stunned to speak.

What I really want is a way to automatically generate movies, so I can finally get that sequel I've always wanted.

10:50 So here's an interesting wrinkle. Gurney is shown painting a house here. Should the architect or builder of the house be able to opt out of their work being featured in Gurney's paintings? My first instinct is to go "no, that's absurd", but Gurney's proposition at the same time sounds pretty reasonable. I'm not 100% sure why.

When I was in grad school, my Computer Vision professor said, Future of Computer Vision tech has already arrived, its the market that is lacking behind. That was in 2020 and every month we had some kind of breakthrough. I can only imagine how it will be in next few years

As an artist, I wanted to give my opinion about it. But because you made a bonus vid. with opinions from artists, I 'll wait and see what they have to say about it. But one thing I've to say about it. The artwork A.I. made isn't art, but the birth of A.I., that is actually the artwork!! In short: A.I. is the artwork, but artwork the A.I. made, isn't. No matter how beautiful it is. But that, that's just my opinion. ツ

This would be an easy way to find painting ideas when I need inspiration but it also scares me that this could potentially decrease painters bc ppl would instead be reliant on ai in terms of making art :/

I have to say that the lighting and composition of the frame in interview of the person at 10:02 was beautiful. Something about the out of focus background, the zoomed in shot, the shadows cast by the trees on their face, and the color of the scene just all work together.

This is the best generative AI text-to-image explanation I've seen. All the designers should watch this.

🎯 Key Takeaways for quick navigation: 00:00 🤖 *Introduction to AI art and image captioning* - AI development in image captioning. - The curiosity to generate images from text prompts. - Early challenges and experiments in generating novel scenes. 01:04 🎨 *Advancements in AI-generated art* - Progress in AI-generated art from 2015 to 2021. - The introduction of DALL-E and DALLE-2 by OpenAI. - The emergence of open-source developers creating text-to-image generators. 04:14 💬 *Prompt engineering and creative interaction with AI models* - The concept of "prompt engineering" in communicating with AI models. - Experimenting with various prompts and creative possibilities. - The ease of accessibility and use of AI-generated art tools. 06:24 🧠 *Understanding how AI models learn and create images* - Explaining how deep learning models learn and recognize images. - Introduction to the latent space and its role in generating images. - The generative process called diffusion in creating images from latent space. 08:27 🌌 *Exploring the multidimensional latent space* - The complexity of multidimensional latent spaces. - How latent space captures different variables and concepts. - The idea that any point in latent space represents a recipe for an image. 10:06 🎨 *Artistic implications and ethical considerations* - How AI can mimic artists' styles through prompts. - The need for transparency in disclosing the use of AI in art creation. - Copyright and ethical concerns related to AI-generated content. 11:36 🌐 *Societal impact and biases in AI-generated content* - Recognizing biases and limitations in AI models. - The diversity and representation issues in AI datasets. - The uncharted territory of AI-generated content's impact on society. 12:36 🚀 *The transformative potential of AI in creativity* - The broader implications of AI in human creativity. - The removal of barriers between ideas and visual content. - The unpredictable long-term consequences of AI in culture and communication. Made with HARPA AI

What an excellent video. The technology is accurately and clearly explained. Truly impressive. Thanks!

This gives me chills, all I keep wondering is how do we know we’re not already living in a world generated by AI?

This reminds me of the nascent days of Photoshop and digital editing. This is going to bring on a whole new level of interesting stuff.

Loved the visual explanation of latent space. Will reference it for newbs. Thanks!