Can DALL·E Generate DOOM?
Вставка
- Опубліковано 13 сер 2022
- Check out Kitten Burst! : store.steampowered.com/app/15...
kittenburst.com
The Kitten Burst Forums Discord: / discord
❈-------------------------------------❈ links ❈----------------------------------------❈
www.jam2go.xyz/
Spotify:open.spotify.com/artist/3p2Ba...
Soundcloud: / jam2go
Bandcamp: jam2go.bandcamp.com/
Twitter: / jam2go - Фільми й анімація
I absolutely agree, publicly sourced AI must always be in the public domain. And we need to start integrating mechanisms that honor image publication rights into scrapers. It's time to make some laws here!
I suspect the library of congress has the resources and skill to catalog "creator rights" on public domain projects, independent of copywrite.
The whole legal sphere of licensing / copyright / patents has utterly knee-capped the West. You can't sue someone in China for 'violating intellectual property laws' in the US, so all it really accomplishes is inhibiting domestic progress and competition, giving foreign actors a clear economic advantage and generating a buttload of revenue for a bunch of greedy lawyers. The origins of copyright are rooted in archaic rabbinical publishing laws and provide zero benefit to a free, modern society. The whole twisted idea of 'intellectual property', 'publication rights' etc, needs to be thrown in the garbage, particularly when it applies to software, algorithms, etc.. they have distorted Western society so much around these ideas, that people actually accept the notion of patenting math equations and shapes.
No
@@roymarshall_ Why?
A little bit of a tangent, but I feel like it resonates with some of your ending thoughts. I feel like the mistake we keep making with internet technology is that we dont form responses to the ethical/legal quandaries before these technologies become available. Because once they’re released, it’s so hard to control the distribution online. I’m also in STEM and I really wish more of us prioritized informing the public to protect themselves from the technologies that our fields create. In the research lab, we get funding based on claims of *hypothetical* benefit to society. And in so many cases, we assume the lawmakers will identify and legislate the gray areas. Too little, too late. The next time we’re hearing about some exciting new tech, raise questions: WHAT is their plan to distribute technology in ways that benefit PUBLIC GOOD. How do we meet our institutions and communities halfway to distribute them ethically? It’s an ongoing battle. Thanks for the video! I wasn’t aware of DALL-E until now
absolutely fascinating. i keep hearing "the artists still have to command the AI generators" but seeing you in action really helped demonstrate the workflow. me using the same tools would not have ended up with the same result. also really appreciate your 3 points at the end, though we know what for-profit tech megacorps will probably decide. the lawsuits will be interesting.
Honestly, the biggest non-ethical thing that bothers me about OpenAI's products is that... they aren't open. I understand charging money for compute costs. But I have a good computer. Let me download DALL-E or GPT-3 and run it myself ffs
I believe you can do that with Stable Diffusion.....I haven't figured it out yet though, I kind of have to go back to square 1 on this tutorial. Totally a thing though, just more fiddly to setup
You _can_ do that, you just need to know how. A friend of mine set it up and upgraded his GPUs to make it run quickly, though I'm not sure if it's the latest version (and he made some modifications, mostly removing the filters).
We don't know exactly what images were used in the DALL-E 400 million image dataset, but there exists an open source approximation called LAION 400m (and soon the follow-up dataset LAION 5b) built from the same common crawl source - you can even search through those datasets online, to see the kind of images used and their descriptions.
I completely agree that using (living?) artists work in an automated transformative way feels... kind of a step in the wrong direction, even if pinpointing what exactly the problem is kinda hard.
Especially your point about what could be done about AI generated art. We could say generating art should have the same bounds as tracing art... but then: what is art? Is a prediction of "I believe the artist X would draw subject Y in this manner" art? If no, then such rules would have no bite to be enforced. But if yes, then what about all the other AI usages that do the same thing, e.g. predicting weather patterns - are they now also art?
The idea that the outputs of AI models should be public domain content is interesting, but likely would also meet a practical hurdle, because as soon as you take the output of such an AI and transform the image by even one pixel, you could claim copyright on that changed image. Also, the concept artists whose jobs will be replaced by AI won't really care if they're being replaced by AI that generates copyrightable images or public domain images.
If anyone is interested in more insights and discussions on this topic, I gotta heavily recommend visiting the discord servers of Midjourney, Dall-E, and various other AI Art services or developers, you can be sure that they have a discussion or philosophy channel, because even the people developing this stuff have no clear picture just *what* they are unleashing upon humanity here.
My 2 cents: Maybe commercializing every aspect of the human experience and aggressively optimizing things that used to be just modes of expressing thoughts and experiences is not a good idea, actually. Still fun before the consequences set in tho.
There's something deeply, bitterly ironic about generating images in the style of Simon Stålenhag's work via AI and tweeting them back at him.
I find that hilarious. Too bad he's a bit too egoistical and seems to be opposing of the technology. Good thing most artists embrace it tho.
@@TheSnufking >oo bad he's a bit too egoistical and seems to be opposing of the technology.
I mean, when a bunch of people are pinging you with "lmao your career's over" with crude mockeries of your art constantly, cant blame him for not having a good impression of it :V
The idea of Dall-E claiming that it can "Copyright" its "art" is the most laughably idiotic thing I've seen in a while. Not only has it been made clear that you CAN'T copyright AI generated art by current laws, but on top of that, you'd be copyrighting something made out of THOUSANDS of other people's copyright art. It's literally stealing bits and pieces from thousands upon thousands of art owned and created by people who have given no consent to it, and then trying to claim it as yours despite it being made of zero original content.
Claiming that you own something just because you made the tools to *steal* something else is obscenely self entitled. It's impressive work, and I love seeing it progress, but anyone who thinks that it shouldn't be Public and SHOULD be paid is genuinely insane. Imagine having your art, made with sweat and tears yet never seeing the light of day, getting stolen for profit and used in some collage. What an absolute joke.
The creators did not *make* any of that art, but thousands of people who will never be credited did. I understand that servers aren't free, and it takes money to grow your AI, but it feels disgusting and greedy to make people pay for a censored art generator made upon the backs of thousands, if not millions, of unnamed artists.
To call Dall-E a collage-maker is a tremendous mischaracterization, just at a technology level.
On a social spiritual level, it is just one of many ways a person can find a mirror for the inner world.
And for many, the courage of holding a pencil firmly is not established.
Can we blame them? The road of traditional art is steep and long, and the rewards are not of this earth.
"A part of you still living on". When you look at it in that perspective these pictures do become a bit more eerie.
Man I love your content so much. Saw the horoscopes on youtubehaiku and thought "these are perfect" and you have not disappointed since
14:49
It's definitely called MS COCO so they didn't have to call it MS COC lol
Really happy to see another in-depth and refreshing take on this. I still always get conflicted with the notion of “public data” on the internet and the question of ownership from media online. Take John Rafman’s “streetview photography” using screenshots from Google Streetview for example. Despite multiple levels of questioning who owns what in those projects (i.e Google or the people in the images), there is still an authenticity attributed to Rafman for the final product that people can purchase. What makes AI so different in comparison to a human using “public data” on the internet to make art?
You continue to be a massive source of inspiration for me. Looking forward to watching the entirety on my bus home w^^w
how do you not have more views this is so incredibly underrated
The issues with using "public" data are very similar to Microsoft's subscription service Co-Pilot AI, which uses all the open source code published on GitHub for an AI model that generates code out of a prompt. Even though the repositories are open, they still have licences (MIT, GPL, etc.) dictating the rights of use - and most importantly, any derivative work (generally) must be under the same licence. It becomes very much a grey area because what licence is the generated code under? What happens if the AI spits out an exact duplicate of some code, somewhere from its training set? What then happens if you use this code in a project with and incompatible licence?
8:57 no wonder you are so good at getting AI to make exactly what you want. The way you describe how the music sheet looks like in four words is crazily fitting!!
21:55 exactly this. I see ui as an inspirational tool. I generate concept art and just spend a long time looking through it for ideas. This is no different then me just going on google images and browsing that.
When you said about AI harming humanity I couldn’t agree more. It’s like almost everyday something new and horrific pops up. Last time I checked there wasn’t ai softwares that could scoop out pictures of random people in everyday normal situations from internet and made them naked and in questionable scenarios (p0****raphic content) but here we are. Regulated ai content is impossible with how internet operates nowadays and it will harm us all at the end. At the end it will be less freedom.
I did not expect to watch all of this video, but I was absolutely hooked! Exciting and interesting stuff, but not presented in the classic UA-cam fake high-energy way. Thanks for a great video :)
As far as I'm concerned, nothing an "AI" (what we have now is closer to a web of algorithms as opposed to anything resembling intelligence) creates will ever be considered art. It cannot create anything without needing to copy something else first. It does not iterate on an idea, it merely takes things other people made that seem similar to that idea and mashes them together into a crude mockery of that original idea. This can never and will never be able to replace even the most basic of human creativity.
At the beginning, was that a modified version of running from evil from doom 2
This has to be one of the coolest and most creative uses of DALLE-2 (and AI generation in general) that I have ever seen hands-down. Amazing!
If you ever want to try something like this again, I recommend Stable Diffusion! It’s almost as good as DALL-E 2, occasionally even better, it’s open source (with the images in the public domain, like you said!), and it can do gore!
Part of my job is in the cybersecurity realm, and I 100% agree that AI use and misuse is a huge issue now, and will only continue to become more significant. I think your simplified "if it used public data, it produces public data" is probably the best distillation of the reasonable solutions to this problem. If you make it so corporations can't monetize the outputs (as easily), it should dramatically reduce the abuse.
Maybe I'm just mad that I haven't found a way to monetize it myself.
Holy shit, that metal music actually shreds!
Also, your insights into the relationship between AI and art is quite thoughtful.
the steel drums in the metal music just killed me
this is so high quality and I hope a lot more people get to see it and learn from it like I did
This was soooo cool :O Makes me wanna try my hand at a similar project
This is awesome! loved the concept and you executed it really well. I'm sure we'll see these tools being used in tripe-A development in due time.
a bit off topic but i never really liked the reaction Miyazaki gave to those programmers showing him what AI can do in terms of novel locomotion animation. He couldn't see beyond the horror themed example they showed him and decided to relate it to a person he knows with cerebral palsy as if that's what the programmers were trying to parody. Like imagine doing all that work and showing it to an artist you admire and he calls it disgusting and inhuman? Felt like a major dick move to me.
What ever is next from you, I got to see it.
Thank you for sharing your brain and your thought process with us. Its a gift.
Honestly I can honestly say that it should only not be public data when the user of that art makes something...transformative out of it, like what you did with the DOOM game. Maybe the assets, like the freako crab monster, the textures and the map, should be public domain, but all of them implemented and modified into a game by a person shouldn't end up in the public domain. AI Art is cool, but it should ALL be public domain works until something transformative is done to the art (something transformative that isn't based in AI, that is, like the reimagining of stalenhag's paintings in DALL-E).
Honestly, making levels and assets for Doom isn't really hard at all, tools like Ultimate Doom Builder are extremely easy to learn and intuitive, and there's a lot of straightforward tutorials on how to use it. It was much harder in the 90s.
Slade lets to edit the contents of .wad files (which is where you put levels and assets for using with Doom), and it's fairly easy if all you want to do is add sprites and custom textures. It's worth playing around with this stuff some more more to see what you can come up with.
Going a bit away from the morallity of the issue, I've heard the argument that "Artists learned from somewhere, and each way of learning would develop different and unique art, the same as growing up, the enviroment affects you and how you become, they study someone elses works and methods, just like an AI learns from a database of patterns to generate something [...] some learn from Van Gogh, and he's not consenting to any of it".
But the thing is exactly the publicity about it, Van Gogh is public domain, but Picasso isn't yet. You can learn from both, study them, but you can't source Picasso to create an derivative work the same way you can with Van Gogh. So there is *technically* a way to only feed an AI with public domain stuff that doesn't have intellectual property anymore and be able to privatize it as a tool for a customer binding it to his property since "copyright in a derivative work covers only the additions, changes, or other new material appearing for the first time in the work", the thing is, in this scenario, when does the gray threshold of creativity, originality and authorship starts when using AI generated stuff?
I think the part about how someone's facial features from a Flickr photo could end up in a Dall-e generated image is a slight misunderstanding of how it works. It's not creating a collage of images that it's trained on, like an automatic photoshop-er. It's more like the ai learns what a person looks like in a general sense and tries to make it's own version from scratch. I feel like this distinction is important when discussing whether ai art can count as plagiarism. But you still make a lot of good points about how it is used to copy specific artstyles and characters it doesn't have the right to. And I do agree with the analogy to tracing.
Exactly, these are novel new images that have never existed in whole or part before, to me it's the same function as to how people make art from everything they've learned before them.
07:38 cruelty squad, nice
power in misery, traversing the grid of death
I too am sad that my username does nothing in DALLE.
You bring up amazing points I never considered. I hate the idea of using this as a way to make art or even make money by selling the art it creates. I see AI generation as a tool. For myself, I use it to help create references for photoshoots I want to do. I can't draw at all, so having a tool like this to put my conceptual ideas into something visual that a model can see is invaluable.
While initially, I thought the clip of Miyazaki was over dramatic, I see where his fear comes from. We already see movies that are devoid of art come out of places like Disney and Pixar, but at least there are still artists making them. While the story and art style is bland, at least some artistic talent is being utilized to make them. Imagine tools like this generating 3d models though, imagine these tools being able to animate whole segments of movies. It feels like it'd be soulless, an algorithm just producing the most mass-enjoyed entertainment possible.
I am of the opinion that if we limit AI from having access to all sources of information that ultimately it will be much easier to feed it harmful data. It's all or nothing.
The fact that DALL-E 2 gives you full commercial usage rights blows my mind. I can't wait to see the inevitable fallout as the tech gets better and it starts to be used in more ways people don't like.
I actually Really like the game you made. I think it looks cool.
Great vid bro!
Every utilized asset should have proper attributions. They should be included alongside what is generated from them. The work should belong to all involved and if attributions cannot be found if using copyrighted material even transformed, it shouldn't be assumed to be now 'owned' by the individual that caused the program to develop it..
How do you get the muffled sound effect at 6:31?
I agree with your conclusions on AI generated art, but I gotta say I straight up hate that miyazaki clip. He was being very unreasonable and self important when presented with procedural animation.
I'm weirdly conflicted with that video which is why I included it. On one hand Miyazaki is incredibly dismissive of digital art, It really was just a procedural animation experiment they didn't mean anything by it. However, they presented it in the worst way possible, and when asked what their goal was they said that they wanted to make AI that can do what artists do, rather than focus on it being a tool that can work with artists.
I kinda get where he was coming from? I think he's saying that an animation of someone in pain like that should have been made by a person who can imagine and empathize with that pain, rather than an unthinking computer. He doesn't like reducing human behavior and emotion to something that can be replicated by a machine. However, humans _are_ machines, and so I think it's inevitable that computers will eventually be able to convincingly fake parts of our behavior. GPT-3, which is really just a word predictor, already shows some capacity for logical and reasoning, simply because most of the text that it was trained on _is_ logical and reasonable.
Miyazaki is unreasonable and self-important in general, but I can see eye to eye with his disdain for this kind of stuff.
@@0neDoomedSpaceMarine But why? Either you as an artist make your own stuff and you have no reason to care, or you as a consumer are met with art that could've come from anything and you have no reason to care. I find it akin to not liking a painting because the guy that made the painting talked shit about music you like. It's an arbitrary dislike that ultimately has nothing to do with the art itself. It's a personal failing to be unable to see something for what it is without all this narrative. The only thing I'd have to say about the miyazaki clip is that the art they showed him wasn't great. But I'd never be offended by the existence of bad or mediocre art, that's the height of arrogance.
@@Jam2go Admittedly I think a lot of my distain for that clip comes from people holding it up as if his word is gospel. I love the work his name is attached to, but he is often a very dismissive and reductive person. The context of their goal brings it more into perspective, but I still think he was wrongly trying to make them feel as though they were being very immoral with what they were presenting. The presenters were obviously excited and respected him, and to respond in the way he did I think is borderline irresponsible. I know that maybe sounds hyperbolic, but you get the feeling he basically crushed those dudes dreams over basically nothing. Then again, I do agree that, at least under capitalism, the goal of wanting AI to be able to produce art all on its own is a very dicey one.
The thing on the payment method is kinda dumb, the greatest cost is the power for the GPUs that generate the image, you don't pay for the art, you pay for the creation of the art
First Solar Punk Game looks Damn good
I agree with you on how AI generation trained on public data should be public domain.
Don't have so much black screen (eg, 8:13) -- people don't like seeing their reflection in the monitor
I love your videos, sir. I want you to know that I place you on the same level as Solar Sands. I can't believe you don't have hundreds of thousands of subscribers, yet. That's a travesty.
Thank you! I love Solar Sands' videos!
Game designers shaking in their boots atm
Remedios Varo is very good and I like that you like her
If you've ever looked at someone else's art then your brain (whether you wanted it to or not) used it in exactly the same way DALL-E does; we should put all your art in the public domain I guess.
"exactly the same way DALL-E does"
@@Jam2go how is it meaningfully different? your eyes receives photons which activate neurons which trains the neural network that is your brain and allows you to generate patterns based on the training data input; the neurons more heavily involved in patterns that get lots of attention generate a reward function that upregulates those neurons and makes them more likely to fire again. you are a deep neural network trained via reinforcement learning; maybe I am missing something
@@Jam2go btw your content is criminally undersubscribed; the birds and geolocating videos were absolutely top tier, like among the best content on youtube
@@owlfrog Thank you!
Sorry for being snippy, I've just had way too many people emailing me about this video and getting really rude about it lol. Yes brains are neural nets, but they don't train or create art in "the exact same way".
DALLE is trained to do a very specific thing, change noisy pixels to look more like what a target language equivalent looks like. It can perfectly recall it's training data and produce a one-to-one copy of an original training image. Brains are magnitudes more complex, as well as imperfect. We don't conceptualize objects and ideas by how much we should change noise to look like something. To say they are the exact same process is reductive.
It's like saying Sims (from The Sims) and Humans experience life in the exact same way, therfore they should have human rights.
@@Jam2go I don't believe memorization is a huge problem with these diffusion models. While technically true that there is a 0.0X% rate of successfully extracting high likeness of the most overfit training images on a purposeful attack, generalizing that these models "produce a one-to-one copy of an original training image" is not a very truthful statement without a huge asterix. A non-zero memorization rate is still a flaw in these models for all sort of reasons - ignoring everything else, it's not even very useful to recreate original training images, but the memorization problem is often greatly exaggerated.
Though it would be very impressive if these diffusion models could recreate perfect copies of the original training images outside edge cases. Would be a impressive compression algorithm cramming billions of images to less than a byte each.
Will kittenburst ever be able to run on MacOS or Linux/SteamDeck?
It runs on Steamdeck! I haven’t tested Linux or Mac myself, but some people have gotten it to work, the performance is best on PC though
Cruelty Squad is just ai generated Doom
at what point does ai go too far
This was very interesting. Loved the breakdown of the dall-e model. Very well explained. In general I think legal restrictions kills creativity (but sometimes with good restrictions opposite is true). As you mentioned in the video everything is a remix. There is even a good video about it - ua-cam.com/video/nJPERZDfyWc/v-deo.html&ab_channel=KirbyFerguson We are all building on the knowledge of our ancestors. Otherwise we would all start in the wild with no clothes on. This would block all human progress. There is only a limited things you can do in your lifetime. I think the way to go is make everything free when you begin and once you start to make some serious money of it then pay some fees to the owners. Like the unreal engine pricing model. This could be applied to a lot of aspects of our lives. Also better education across the globe. So we don't have yokels taging Simon Stalenhag with BS. By the way, one tweet from Simon. Even though he clearly does not like the AI generation of images or maybe just the way people are approaching it. "If we would sue every time one artist's idea shows up in another artist's work we would do nothing but sue each other all day. And I truly believe we produce lesser art if we constantly examine and censor our work based on legal anxieties."
You should've used Stable Diffusion
you look exactly like nick robinson in the thumbnail lol
wheres the doom gameplay
why do you want me to do you?
Amusing narrative... good laugh.
The one thing I think you have gotten totally backward is the idea that anything 'created by AI' should be owned publicly. It seems like a notion that has a real possibility of being very popular, however, it runs into multiple problems when you bring it out of the snapshot you have chosen to emphasize.
Firstly, the careful choice of words to evoke imagery is an emerging art form. 'created by AI' is a misnomer.
It is foolish to say that we who evoke said imagery are lacking artistic merit.
You might make a similar argument about photography not being real art because it doesn't require the skill of a painter... but that is a problem that only exists for people who can't tell the difference between one and the other. It is a product of this tiny snapshot in history, and not intrinsic to the medium.
To drive home the point, if you can't tell the difference between AI produced art and real art, how can you possibly exert ownership over these things you would like to be 'public domain'?.. and if you CAN tell the difference, why are you trying to exert power to subvert this new medium?
Thank you for the thoughtful response! To clarify, my main point wasn't "anything created with AI should be public" but "anything created with AI, which was trained using public data, shouldn't automatically grant ownership".
Working with AI does have artistic merit, but it depends on how you use it. Take your photography example, does taking a photo of a painting have the same artistic merit as the painting itself? No, that's just using photography to copy something. Artistic merit in photography comes from using the medium to create something new.
My main fear isn't artists using AI as an artistic medium, but companies and grifters using it as tool to exploit existing art.
My goal isn't to subvert AI as a medium altogether, but this specific application of it.
I think viewing AI as a form of tracing is the best way to put it. You can use tracing as an artistic medium, but you can also use it to rip other people off.
I think the baseline for AI generated art ( using public data ) should be public, to prevent this exploitative use case. However, if your work is transformative enough ( like with tracing ) you can justify your work as being original. To me that justification is important, you shouldn't automatically have the rights to something because an AI made it.
@@Jam2go Totally spot on.
Great point about a photo of a painting. Thanks for taking the time to read and respond 🙏.
Btw I didn't hear any mention of Disco Diffusion and Stable Diffusion in your vid. Both are open source, though they have the same issue of scraping public data without any compensation to data contributers.
Disco is much more expressive than Dall-E, and Stable Diffusion (newest) allows for producing imagery without openAI style censorship.
@@AZTECMAN I haven't played around with those yet! I'll have to check them out
That David O'Reily is a massive whiner. It'd be a scam if they literally used somebody's pics and sold em. This is like looking at pics, remembering them and redrawing them. And this whole apparatus - to just redraw pics - costed a billion dollars. People like that drag down the progress of human civilzation.
Okay, I haven't finished watching the video but I just want to put it out there that at 1:00 @Jam2go you and I have that exact same shirt and I just think thats really neat