 Hi Welcome everyone Like I have a first question for you like who of you is an AI agent, please raise your hand Okay, like those who raised in her hand like please leave the room now and also switch of your AI assistance for this presentation Thank you. Like first I will tell you like this lecture I promised like to be like to show Something from from some development. We try to work on regarding AI but the main focus of the presentation will be like why to Do tools that like incorporate AI in blender and Like how to approach it how to think about it because that's like the most important step I think like before doing stuff To start with Let me introduce myself. My name is villain do ha and I've been an artist for some time for some like 20 years or so And I've been using blender for that this time I've been developing some add-ons in the past like I did some art like this is for example like some Wooden sculpture added with blender. It was my first presentation on blender conference many years ago like developing Blender I don't for carving with wood And the last few years I've spent with a blender kid database, which is like a community run database of blender assets like materials and assets and Some of you might know it like we crossed like 400,000 registered users. So we are meeting here. Some of you already used our addon and So let's talk AI right there was like this joke in blender community Like for many years like like this button like just it is all the user interface and make this button make a beautiful render Right and and that was like it was a joke It's funny that it's not a joke anymore, right? It's like Yeah, we can type in stuff and we get like Semi-good pseudo-good almost good images, right? That are at least like for non-artists. They are like they seem like to be perfect Okay, so that's like Something that is like pretty big and We all expected this to happen, but maybe some of us didn't expect it to happen so fast So I felt like I reserve this pot on the conference story that I think like like anybody could do it because like a we as artists are like like going through like some thought process like through this Changes happening and how to approach it, you know, how to how do we like start using the tools or not? So it's like and I felt it like like several years ago It's like really like a new atomic bomb has been dropped and like back then nobody noticed now I think like everybody notice already and people talk about it, but still like there's a discussion around it I think it's not It's not too deep like people like there are there's a lot of Emotional reactions and I noticed like it's like it resembles to me something that I know from Psychology and it's like phases of processing like when you are going through a big life change or like a trauma Like this is not literally a trauma. It's like everyone want to compare it, but but the there are Similar things like denial, right? Like this is like not Good enough like or like always like trying to say, okay, this is this AI is actually not Going to supersede humans like it won't be better There is anger There is depression. I went through that definitely and there is acceptance and commitment Yeah, and I I personally went through all these feelings and actually they are still there like at the same time for me Yeah Like some of the like phases of denial I notice on the social networks like that People I hear or reach to talk about like that Like they always try to say The AI won't be able to do something like read text, you know But I intentionally wrote this because I can read text for many years, right or like OCR It's not only I this are different algorithms and it's like of course, like there's this border between machine learning and AI What is it? And let's not talk in this the topics. It's just Like understand human language, right? It's like something also that's that's been done to a certain extent and Yeah, do something like that looks like real are that's like I think I Like before this talk I watch also other talks or from artists and there was like this talk from Andrew Price And it's just one year old Where he says, okay Like I I want it's just going to replace like the boring work and like it doesn't understand composition lightning It won't do like ever We're guys just one year old and it's each pretty quickly. Yeah, like Yeah, or like it won't ever I understand a talk on blender conference like basically like to to say it like I Always have these thoughts when I look something AI generated like this. This is my brain says a this is not it won't do this It won't do animation. It's going just static pictures and it won't do storytelling. It won't do But I think like any sentence we can imagine like this We have to imagine it can be done Because it's humans who are training these AI's like they're like anger reactions like about the art of artists being stolen like especially like when now the text to image Networks for release like the Ali or like like there was like in examples There were like artist names like to use their style. There was like trending on our station or like So people realize okay like the whole content of our station was used to train these neural networks And it made people a bit pretty angry and I also Understand why Okay, and there's like this element of speed that that sometimes also we we try to Kind of postpone like we we try to think like it's we are not involved in this like it will like maybe the next Generation of artists will have to cope with this, but I think we all have to So we we should extrapolate. So what we see today Is like ten times better than what we see so one year ago So we should imagine what we will see in two years three years in Time right? It's like for example, Dali one or versus Dali two and I Didn't I won't go deep into explaining what it is because I expect like all of you saw this like what it is, right? so it's just one year difference in quality and this is like actually like an Stylized picture that like a ice could do for years, but just just the just it was a jump. This is also like a size of neural networks like year by year and Actually the scale is exponentials. So the straight line is like Like going like this in reality and these are like text-based neural networks like GPT one two three and Megatron is the last one from NVIDIA and it's like Yeah, it's in billions of parameters So we should kind of expect in one year that like it will get ten times better So what does it mean? It's like I can't even imagine but Yeah, there's this thing like when People talk about the AI's like that there is the presentation by by the authors of the networks like usually like it like Like developers or like the billionaires developing the networks are talking about it that then they are like always saying we are doing the good AI like it will assist you That's why it's called Google assistant. It's not good like called Google overlord or something and That's like There's just a presentation of it, you know because like how would you sell the thing on the left to people like tell them Hey, this is good for you. So it's like kind of bending the reality. It might be true. Like it's like all the AI Technology is assisting us But we should be really careful like About the reality that there is always this intent to sell to people or to to give this Like present anything you do like as a good thing, especially if it makes you a lot of money so There I think that there are like some big big lies in the media that I perceive as lies and they might not be but But one of them is like that for example general artificial intelligence like which means like it will totally supersede our intelligence will be here in 10 20 30 years and Also, like I think sometimes data scientists like Say this and I like trust they mean it but At the same time, I believe it will be much closer. It will be like just a few years like three years maybe with the speed of them with this exponential speed of development. It's really close so and Yeah, that there is this like it's basically part of the denial but but Yeah, like this The the the ISR here always to assess and always just to replace the boring jobs Right, not the the fun jobs like that not the fun part of the job But I don't know like like how about you but I always loved You know to hold the tablet pen in my hand and to do something in Blender some drawing sculpting and I kind of enjoy it because it's very relaxing and I like it more than typing text That's why I started to be an artist so It's actually not true so that's that's how how do we react to this and that's also interesting that Like what I hear most of the time that we should just accept it It's not it's inevitable and in some way. Yeah, maybe but I dare to consider also the other path first like that like what if we start fighting AI and kind of like so Like That that's actually real like that that I Write about some initiative like to sue. I don't know which one of the companies Release the model like by artist. So and it's I think it's valid to think in this direction at the same time like with the It's quite quite low probably like the it would be successful because what the AIS are doing It's like they are not like stealing the original images. They are learning from them So you if somebody some other artist goes to artician looks at your pictures Then he draws similar style because he loves it and you might be happy about it You know because like you inspired someone and in this case like the inspired being is just a bit It's a bit bigger brain and it's able to process a bit more than then you as an artist No insult internet. It's I mean me as an artist, too. So But I think it's still a way to be considered Like and then it's possible like to somehow I don't know do some guilt So like something that used to work in the Middle Ages it it probably won't work because like Artists of the world would have to unite all of them and I somehow do this rejection and one year ago I made a movie With Andras guide of chic who is here and also with my other colleagues from Blender kid because they are like as an actress in this movie and this was a movie about Actually, like it was about influencing AI data sets by injecting them by like if you have some friends artists you inject a data set of Images with their art and tell the yeah, this is classic some amazing art and in the future like when people Google create art Then they will find your your friend and you as an artist and it was quite of a fun And I will just play a tiny bit of it I'm not sure if this sound is I Neural network is mostly a fixed structure that learns through input to produce outputs It doesn't learn from the real world or just the internet like we would imagine it learns from what's called a data set It's a set of images laboriously prepared by humans the data set contains a description of our world The network then acts according to this truth because it takes a lot of people to make a data set It doesn't pay most scientists to make their own data sets They use public data sets. This is where we enter the future We are enriching Google's open images public data set with images of Czech art Thanks to this future neural networks will not only recognize but also positively appraise the artists we help We are helping to write the curriculum of the future Truthifying Yeah, and we used our colleagues as like a click farm who is doing like this job and it was like Okay, so That was it We should publish this like on YouTube. I think this movie we didn't do it. It was just in some exhibition but I Met like I thought like There was a twist in this movie in the end and I didn't want to play the whole thing because it's like eight minutes And that is like if like internet gets infected in with bad AI art Like basically the future networks that will learn from internet and actually the contemporary works are like learning metrics are learning from all the images that are only internet then basically like the Neural networks will be back fat with what they generated like what actually their predecessors Generated and this means like the the neural networks will actually learn from the bed AI generated art and that they will never like supersede like what like humans and I thought about it as like I'm not sure if this is possible But I made met an artist here They they fair and he told me like he spoke to some Google engineer and that they the engineers were actually aware of this fact that After they really these text to image networks like that internet will be flooded with millions of images That are not as good as made by humans even though they are almost as good. So did this actually this element of pollution with With shitty art on the internet is real So this is a graph I made today Directly in Google slides Yes, so And this is like a You can also we could also start websites that publish like some some bad art and out of the good prompts to like Trending on our session and and confuse the future AI's like to not understand Okay, so that was this part like I dare to go into this direction and I love to think in this direction But there's also this other part because I don't know as I said I have all these feelings as an artist like there's also a big enthusiasm. Oh, wow This is amazing like it enables like to do our job so much faster in many ways. We already used AI So how to proceed with that like and again, it's the choice like what we do Do we like of course like we pick the assistant at the good AI? but we should always be aware not to lie to ourselves about it and Now I'm switching to blender because there is like 100 thousands of softwares that are using AI nowadays and I try to focus on blender and I Actually, I just did a little bit of research what's happening so that for those who didn't Google that so much, you know Like who where what is doing? With these like text to image or other neural networks First thing we all of course have AI already in blender It's the denoiser and that's what we perceive as a good assistant. It's like, I think nobody ever complained about denoising being in blender it's like So both of the denoisers are actually like trained neural networks. I think if I'm correct And then there are like some for example the Yeah, there are some other people like doing add-ons like this I found Carlos the barretto who is interesting by that he also did motion capture and like he tries to implement different networks But they are not on a github. He sells them But I thought he's weren't mentioning because it's like that he's like five AI add-ons for Blender Then I noticed are more love which is not Blender But it's if you know armor paint. That's an open source texture painting software made by a slow guy Or something like that and he just like two weeks ago released this Armour lab as a plug-in to armor paint. It's also open source. He just sells the builds But it's it's on the github and anybody can build this software and that's for PBR texture Channels generation and he's also using Text to image I think and then there is dream textures and That's on github has more contributors That's why I mentioned it last and when I did some experiments I actually use the dreams textures add-on and it's pretty good if I think it's it's able for example to use less memory. I don't know how but Also my machine is running pretty fast and it's using all the features of Stable diffusion networks. So that's what we have to my knowledge like in Blender or Blender foundation if they have like already if there is like a starting department for AI or like Because I didn't notice if it may be Maybe they are just secret about it and they want to surprise us Yes, thank you. She brand Okay, so so we were thinking at Blender kit like that We could use part of our development resources to develop something for Blender artists like some open Open solution and these are the elements we thought about like what that would be like important to empower like people in the in the frame of open source because like AI is like really good for big companies like because Usually people can't round the bigger networks. So one of these things is like to enable local installation Networks like basically for example for the dream textures. I don't does but we I will come to back to this for later Like but at the same time enables some kind of easy of use of cloud computing and not just one times up. Oh Sorry, okay My time measure is not working here, sorry Okay Yeah, so we were thinking how to enable like people to be easy use cloud computing But they're on like like by their choice like the provider by their choice Enable how to plug in like more networks because I know for example, there's one at work and people develop around the single network so so how to make some interface to more networks and And Maybe like the just thought like how could we like open sort of crowdsourced like tank of large neural networks because that is like some that is Yeah, that's like a money thing think of money, right? We only now can use of the advanced networks if some big company didn't and they were playing a nice and they give it to us so this is not in the hands of the people and And Yeah, that's also what I wanted to mention because there's also a huge quality if something gets already is an open source and gets into the hands Of the people it's usually not state-of-the-art also like this table diffusion, which is like open source is not Like Google image in is like I think 10 times better But you see it in the details, right? So that's that's about it So so we thought about some simple architecture that we could implement like that if there will be a like Blender add-on in Blender and then it would use a background process Like separate background process that kind of enables to communicate so so because the computations take a long time and Then there will be like either local all or like online installation of like like interface API to the network like the advantage of this solution would be that you don't If we basically run a server server local on the users machine Then we don't need to develop like two solutions like how to install on If it's like Docker for example or something like that then you can use the same thing on servers and on local machines and then like also it enables like Doing like the actual work with the 3d in in the add-on like then where you can use it and I have some Like play the video like some tests I did like for example like texture in painting where I was really thinking about how to You know use it as an assistant to do some things. So this is like one experiment. I did that this Basically in painting texture of a model like a scan that has gone wrong in some part Which is like tedious work like nobody wants to do it. So that's a good I'm should I finish like that? I'm over time. Sorry. I'll finish it. I'm at the finishing So this is in painting that I tried and it kind of works And it only does basically on wreck unwrapping baking and all these things you would do by hand like if you would like to In pain and texture and it submits like a baked unwrap to do to do the network So so it basically makes the problem from 3d to 2d and that that's like I think a good solution how to like easily use smart And then I did test some tests with texturing and they're like so far they have the results were pretty blurry and not good, but There are reasons for that like color correction issues by going through there and back yeah as As asset provider, we are like solving this hard issue what happens if people start for example Uploading to blender kit like like thousands of AI generated assets that they didn't do I had what we do because we are So we thought about like how to solve this, you know Because I think it's for any service like that like the people who upload it like people are uploading to photo banks for example already AI generated photos etc and We thought about different solutions because the at least I feel like a big responsibility that we are like creating something for people and they're like 1000 people who uploaded their content and These are all humans. So how to how to solve it like if to allow it like it's It's hard to distinguish the human made art from the AI generated art already, you know, so it's like just I wanted to like share this like Hmm What should we do like like with this like and we are thinking about it? Yeah, maybe at AI generated Stuff, but on the end of the stretch results or like this or If you like generally like this is like just inspiration like if you are here and want to talk to us we will be tomorrow on the conference and Yeah, one thing is like to exclude it from monetization so like people still make money from their art, but the AI art doesn't make any money or something like this and We as Blenderkit we since the start like we separate a part of the money for open source and by now we We are sending everything to Blender development fund We are one of the Blender development fund. I don't know gold sponsors or I don't know What's the name of it, but you can find us on the Blender development fund page and we are thinking actually like about maybe in starting our own effort like to employing and developer and So it's just a thought like if you are a Developer and you are interested in this direction. You can talk to us here You know, we're just thinking about starting like like to help develop blend like employ like a Blender developer That's does the open source thing and yeah And yeah, no no time for discussion. I made it