NVIDIA’s New AI: Generating 3D Models!
ฝัง
- เผยแพร่เมื่อ 19 มิ.ย. 2024
- ❤️ Check out Runway and try it for free here: runwayml.com/papers/
Use the code TWOMINUTE at checkout to get 10% off!
📝 The paper "GET3D: A Generative Model of High Quality 3D Textured Shapes Learned from Images" is available here:
nv-tlabs.github.io/GET3D/
📝 Our paper "Gaussian Material Synthesis" is available here:
users.cg.tuwien.ac.at/zsolnai...
Andrew Price’s legendary donut tutorial: • Blender 3.0 Beginner T...
🙏 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, Edward Unthank, Eric Martel, Geronimo Moralez, Gordon Child, Jace O'Brien, Jack Lukic, John Le, Jonas, Jonathan, Kenneth Davis, Klaus Busse, Kyle Davis, Lorin Atzberger, Lukas Biewald, Luke Dominique Warner, Matthew Allen Fisher, Matthew Valle, Michael Albrecht, Michael Tedder, Nevin Spoljaric, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Rajarshi Nigam, Ramsey Elbasheer, Richard Sundvall, 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 background image credit: Rockstar Games
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Chapters:
0:00 Intro
0:31 1 - Great meshes
1:32 2 - Variant generation
2:40 3 - Textures vs geometry
3:11 4 - Text-guided generation
3:28 VS previous technique
4:50 It gets even better!
Károly Zsolnai-Fehér's links:
Instagram: / twominutepapers
Twitter: / twominutepapers
Web: cg.tuwien.ac.at/~zsolnai/
#nvidia - วิทยาศาสตร์และเทคโนโลยี
Here's a list of all the parts of a video game, and the AI's that can generate them:
Code - OpenAI Codex
Sound effects - AudioGen (by Meta)
Voice acting - Uberduck, RAD-TTS
Music - AIVA
Textures - Stable Diffusion, DALL-E 2 (Maybe midjourney?)
User interface elements - Midjourney, Stable Diffusion, DALL-E 2
3D models - GET3D
2D graphics - Midjourney, Stable Diffusion, DALL-E 2
Animations - ASE, AI4Animation
Scripts for characters and dialogue - GPT-3
Level designs - (Maybe Midjourney, Stable Diffusion, DALL-E 2?)
(I need help refining this list, please reply with any better AI's you know of, or parts of a game I am missing :D)
For voice acting, I think RAD-TTS by NVidia is better (th-cam.com/video/RknIx6XmffA/w-d-xo.html ), but the caveat is that it is not user friendly (you need to know programming and their documentation has missing steps). I figured it out eventually but it took a while. Hadn't heard of Uberduck before, their front page Eminem voice is good but TH-cam videos show that it isn't typically that good and still slightly robotic sounding.
For music, AIVA fits better for game music imo ( th-cam.com/video/QhyTokHi-pI/w-d-xo.html ). The AI only gives the notes and doesn't play it like Musenet but it has a built in synthesizer to play or edit it.
For animation AI4Animation is probably the best but I think it is closed source as I think EA bought them. Otherwise ASE by Nvidia ( th-cam.com/video/1kV-rZZw50Q/w-d-xo.html ) for physically simulated characters looks like a good alternative.
@@13lacle That's great, I've added those to the list. That AIVA song sounds incredible!
yeah I really think we are less than 10 years away from a "make game" button. it may be like holodeck star trek TNG episodes. "computer, make me a relaxing 19th century romance adventure that takes place in a dystopian steampunk London, and it stars Beethoven as the police chief and a mix of Donald Glover and Colin Farrell as the wealthy french nobleman suitor. with a little bit of action and espionage, that lasts about 3 hours."
Bruh all my time spent learning blender, music, art and coding to make my stupid games, and now you can do this? God damn dude, this is sick but also terrifying. Although, I gotta check how clean the geometry of the AI generated 3D models are, but if they are good enough, gone are the days of having to spend 2 hours just to make a high-res fully baked rock model.
@@conditionalbee9603 its not that good you know .... this means you will have a an oversaturation of gaames to this point that it will have no value to humans no more ...
Geesh if I wait long enough, AI will do everything that I planned to spend time doing
Frees you up to focus more on the big ideas :D
@@ninjuhdelic more likely humans will just be consuming ai-generated entertainment like zombies 24/7 like in wall-e
@@ninjuhdelic copium
@@ninjuhdelic AI will do these too
you can still do them. I drew some titties the other day and I enjoyed the process.
2 minute papers from 2 minutes ago? 😌👌
SLAYERRRR
Oh my, I've never had a pinned comment before!
@@sab3r10 Prepare for a barrage of notifications.
the best kind
Hold on to your papers!
4:19 “The AI does all the hard work, and we take credit for it”
…sounds like a motto of the channel
What is the point of making a tool if you not gonna use it for your benefit ?
@@fakerobot0017 Imagine you study for your whole life to become a doctor. Then one day, someone released a backpack that prevents and cures all health problems known to mankind. Your job, your life - was just rendered useless in an instant. And while some might say "but that's an extreme", well - no - it's a reality - journalism, is already suffering at the hands of GPT, we're seeing more use of it in public realm from senators using it to generate speeches to organisations like the BBC flat out using it in their video broadcasts. We've seen how artists are affected by image generation, and musicians are coined with creating new songs they didn't make. It only takes two seconds to see the implications of these tools, and of course - it only takes two seconds for a bad actor to use these tools to do some real damage to someone's reputation, career & life.
That's a ton of 3D artist jobs that are disappearing here. Illustrators, concept artists, portrait artists, gamedev artists, vfx artists, musicians, writers, ... All on the chopping block already. What's next to become obsolete I wonder? And I can't help but wonder what all those creatives are going to do next? All rush to the few remaining creative pursuits that can still earn a living? Give up entirely and flood the manual labour jobs, thereby pushing wages down? I feel like a lot of us vastly underestimated AI. Most of us thought AI would replace the shitty jobs and free humans to do creative work and innovate. Now we're seeing the exact opposite of that. It's so soul crushing ...
As a 3d artist professionally this is really scary lol
Agree cause I’ve been teaching myself 3D modeling for the last year and now feel that it was almost pointless lol 😅
AI scientists don't care about you or your job unfortunately.
Don’t worry. AI can’t yet make efficient topology or maintain a good polycount
The scary thing is that you feel scare by that. What. You could just find another job. As long as the world have problems there will be plenty of jobs everywhere. Someone have to solve them.
@@jspr49 for now.
There will come a point where computer games are fully AI driven
matrix bruv
I thought this might take many decades but it could be as soon as 3 to 4 years if that.
Endless path generation in metaverse, finally.
Imagine typing about 1000 words of parameters and the AI generates the whole world, characters and story with that. But the gaming market could instantly be oversaturated with a bunch of games and we wouldn't really appreciate good games anymore because the options would be infinite. We'd probably get to the point of quantity over quality. It's scary but exciting at the same time.
I can imagine a future where installing games is just downloading a 1GB AI instruction file and the rest of the game gets generated on your pc
"the ai does all the work - and we take credit for it" -- welcome to 2025
Well before that it was the workers who did all the work and before that it was the slaves. So uhm progress :D?
@@prometheus9096 The workers were paid for that work. If AI does all the work, you dont need workers.
@@Veeger and the slaves were dude?, do you even read what i wrote?
@@prometheus9096What? I dont think you've got his point. dude.
Now you know how God feels ;)
/Joking
But still, think about it🖖🏻
Or is it, humans does all the work, and then God takes credit for it🤔🧐
Doctors and staff "Your daughter will survive"
Parents "Thank God!"
I was listening to some historic recordings of pianists from the 19th Century, very poor audio quality of course, and I was wondering. If you fed an AI the score of the piece being played, and the recording, perhaps the qualities of the instrument and other metrics, could it generate a faithful recreation as is done with video? To hear an 1889 performance of Brahms in HD audio would be quite amazing!
Have the AI read and play all the manuscripts on IMSLP. In any style of pianist (or singer ), you desire! There are millions of pieces of music and only a thousand or so get performed in public.
I want to hear what old timey radio voice people actually sounded like
I was trying to get my audio engineer friend to work with me on producing parallel training data for an ai audio fidelity upscaler but..
Musicians man, musicians.
@@encyclopath A lot of actors and musicians who were young in the 20s (when the first electronic recordings were made) made their last recordings during the high fidelity 70s, which was the height of analogue recording technology (Queen and Pink Floyd). As such they left recordings across the full history of colorful vintage recording set ups. With Spotify and You Tube, it's easy and fun to explore the answer to this question right now. And an AI is only an attempt to model the auditory processing that goes on in your brain anyway. Anyway, for a musician check out Louis Armstrong, he's the all time greatest. For a speaker, Orson Wells is an iconic voice who recorded in a, to a certain extent, a hilariously broad variety of contents across the 20th century including some of the greatest films of all times.
@@adamedmour9704 I’m more talking about speaking voices from ww2 newsreels, radio broadcasts and voiceover narration, rather than musical vocal performances, but that’s probably something I could investigate similarly
To the guy who said in the last video: "It's going to take decades before it starts replacing jobs": Are you sure?
as cool as this is, as someone who enjoys the process making 3d models and wants to provide that skill to others, this does worry me a little
AI scientists seem hell bent on ruining the ability for anyone to make money from the creative arts. What problem exactly are they solving with this crap? Can they not focus on solving social issues instead of causing them?
@@macjonald Well on one hand, it can enable people to create freely without relying on others, or spending time and money they don't have to invest in creative skills. Which is great because it can encourage people to make the things they've always wanted to without feeling restricted. But on the other hand i cant imagine too many people would rather hire an expensive artist when they can just have the work done for free in far less time. Unless commercial use of the ai is wildly expensive as well, but you still get a result faster. its kinda a double edged sword. they're fixing a problem with a problem. they should be using ai to refine the tools not the craft itself
@@macjonald they usually say that it’s not a substitution of creative people but just a useful tool or helper 😅 But eventually you see that AI can easily replace a human being in many areas.
There will be a day soon where you can build your own 3d model from scratch and use AI to customize it, modify it and even create design variations from it.
The way I'm seeing it, it gets you 80% there, it lays down the basics and it's up to you to go in and make the 3D model truly great, amazing, thought out, and make sure everything happens in the way you envisioned.
This is really interesting and cool but scary. We are all gonna be out of jobs in 10 years at this rate. We all thought AI would automate manual labor first and us humans can spend our days enjoying life and making art.
But now they are trying to replace artists first. Lots of people rely on art for income and it is one of the most common careers for disabled people.
Sure people can still make art without AI if they want to but if we cant rely on that for food and housing then thats scary.
We as a society have not fixed food production, energy and resource crisis and implemented a moneyless society or atleast a Universal Basic Income but we are already replacing art before other jobs or making the concept of work moot.
hard to imagine the world a decade from now
There won’t be much difference a decade from now. You might see more poverty though
@@youtubedeletedmyaccountlma2263 Are you sure about that?
With AI alone the late 2020s will be vastly different from the 2010s, and we are not even in the 2030s yet...Change is coming, and we better get ready (if we can even be "ready").
@@denisgabriel4645 yeah. The world has always been like this. In many part of the world, the poverty continues, underdeveloped, cycles continue
@@youtubedeletedmyaccountlma2263 I don’t get from which perspective you’re seeing the world.
But to me, the world is already way different and better than how it was centuries ago, and it’s dramatically changing as we speak.
now you are in the same position as people in 1999
After coming for 2D Artists now they're after 3D.
As a 2D artist, I think it's better that it comes for way more jobs. Then maybe people start waking up. 2D artists alone are too weak (no army of lawyers, like the record industry) and are silenced by all the "Get a real job then" idiots
wonder when it goes for 4D artists
It's good because, i think help new developers and artist, such as new indie devs abd such to make their first projects faster
@@shredd5705 get another job
You must see these tools as a calculator for a mathematician. If you see them as a treat, you would be better off in a retire house with other grumpy old men.
I can't believe how many papers "Ours" and "et al" have done together
and so many different fields! From Computer Science over chemistry up to quantum physics! Why didnt that guy won a nobel price?
Or all the songs Feat sings in.
Wow. That's pretty incredible. The next 5 years are going to be insane.
5 months
or 2 months
or 2 minutes
Not for me
It's basically the end of my job / passion
@@EddyBF well, the 3d models still look like shit and stylization is definitely still pretty far away. probably a solid 10 years before you have to worry assuming you're a 3d modeler.
Clearly AI development greatly accelerated the last few months, maybe I'm wrong but I feel like ever since the success of GPT3 it seems many advancement happened from refining the technique that made gpt3
Language model-based AI's got us so far and still seem to be scaling well and solving new problems weekly.
Some Google AI engineer said after GPT-3: "AI will be able to solve any problem that can be described by language". (Not an exact quote)
There is still no AI yet, it's just a prediction based on the training data.
But this immitation of an intelligence is getting more and more convincing, yes.
@@jendabekCZ "Prediction based on training data" is how intelligence generally works, is it not? Who's to say intelligence or even consciousness doesn't arise from any complex enough network of connections? Our AI's are still meagre compared to a complexity of a single brain, and it's already showing vast potential.
@@z3dar Scientists say a human brain works differently, as it can work with context, estimating things, improvization. A child doesn't need to see 1000 images of a cat to be able to identify it.
In contrary the current "AI" algorithms are always specialized on handling single tasks. An AI which is trained for Starcraft 1 can't even play Starcraft 2.
But yes, even they admit nobody knows whether even this kind of progress somehow can't lead into a real AI.
@@jendabekCZ What does it matter though? Why do you need a single generalizable algorithm when in any practical situation a specialized one will work fine? Sure it will be awesome to one day have AI/AGI that can learn and do any task but is that really necessary when the specialized ML models we have now are giving us lots of progress
The growth at which AI is growing is genuinely scary,you may say that it won't matter in a large scale but think a little bit.
10 years from now,the whole hemisphere of artists and 3D model designers is gonna look drastically different.
Entry level market is already over saturated.... it's gonna only get tougher and tougher for them to break into and get a stable job.
We should have solutions to both accelerate automation and help those who's jobs were replaced. Hopefully we adopt some form of UBI and find a solution to prevent wealthy individuals from having so much power i.e.:Owners of large businesses like Walmart.
the implementation of a Universal Basic Income will become necessary
@@itskittyme
Where would the money come from, and is it enough?
@@itskittyme I think a new currency system would have to be in place for any country to do this. Plus who knows how much money your getting as basic income
@@bengsynthmusic Imo money would come from the government the same way it does now however if what you meant was "how would the money circulate" then I would imagen a system of taxation would be placed to help pay the UBI. A fully automated economy would probably have a 100% tax on the robots, in other words, when you buy something made from a robot you're just paying yourself and in this system UBI's purpose is to limit total consumption so that demand doesn't overtake supply(very similar to how money and markets work right now).
That's actually quietly much more impressive than it probably first appears.
I wonder how they achieved that kind of functional completeness (like car doors latent space interpolation while remaining functionally complete)??
Every adversarial image generator I know can't do this (just look carefully at hands - they are almost never feature accurate).
As a game Dev, artist and ML enthusiast, I actually wasn't expecting this for a while. There are actually a lot of not very obvious hurdles. Very impressed (and a little scared lol).
We cannot anticipate exponential things.
What you think may take a couple yrs likely will be done by mid next year.
"...and you don't even need to be a professional artist!"
"We just type and hit Enter, and we take credit for it!"
Yeah that about sums it up
You're an artist aren't you? Sounds about right
This is truly remarkable, how long does it take the algorithm to render these models and their variations? Hours or seconds?
That'd be down to the capability of your pc and graphics card.
@@Veeger seems "runway" is browser only, plus wonderful "pay monthly forever"
it's unironically looking grim for digital artists
Just graduated recently as a 3D artist, looks like it's mum's basement forever for me
@@Merkygloom I feel your pain. :(
I'm truly scared about how disruptive this technology is.
??
@@nicreven it will undoubtedly be used by companies to get around having to pay artists
@@MishaHivemind Good. This way games can be made better and faster
@@MishaHivemind I hope ever job gets automated.
@@MishaHivemind i don't really think that's ever gonna happen to be perfectly honest with you
They'll get good. Really good.
But there are still reasons you'd want real people working on these things
This is absolutely insane. As a musician, gamer, and an oldschool html programmer, this has sparked some interest for me with what could be done with this technology.
Pretty soon these models will be taking from a huge library of high-quality STL files, allowing AI to edit, mold, and shape them using guide curves and prompts. It is important for 3D artists to carefully consider the copyright clauses in their models to protect against unauthorized use in AI datasets, or pull them altogether out of your online shop while you still have time. There is somebody right now as we speak purchasing your model for use in a dataset, in fact Laion already has a library of 30k models from thingiverse for use in their dataset. Let The current situation with 2D artists and the theft or unauthorized use of their work serve as a warning to 3D artists. Fortunately, we have time to prepare, as it will take longer for AI to be able to effectively manipulate 3D models due to the limited availability of freely accessible, high-quality 3D models. Take my word, what is happening now to 2d artists will happen to 3d artists as well pretty soon.
This is worrisome
Other than copyright concerns, you're cool with the technology? Or do you have other concerns, for example that they might "take our jerbs"?
@@Jay_in_Japan not sure they will take our jobs, even with the Ai, there will still be humans needed to art direct and to assist. Other than that, I only have positive outlook for this technology as long as they don't get there by stealing from 3d artists works they created with their own hard work.
@@Jay_in_Japan I don't even care about the copyright concerns. No one actually gives a damn about copyright unless it effects them. Art and knowledge should belong to everyone. The issue we really need to have is an issue with capitalism.
@@EmpressEris Art and knowledge should belong to everyone ? Sure but who is gonna pay the artists that create that art so they can survive ? Imagine next time instead of getting a paycheck your boss goes "free labor should belong to everyone" and refuses to pay you.
And I was recently thinking, 'yeah I got nothing to worry about for a while, I'll continue learning Blender'.
Here we are though, maybe Ted was right.
Hey two minute . How about sharing your thoughts about the fact that these technologies can very well end up destroying human jobs . What are your thoughts ? Do you think AI will completely dominate the industry or will humans work alongside Ai?
My thoughts...
Sign makers > 3D printing
Type setters > Computer text
Printers > Desktop publishing
Illustrators > Photography
Darkroom > Photoshop
Filmmaking > Video production
Live band > Recording artists
Audio techs > Digital audio workstations
This proxy parallel list goes on and on, and that's just creative arts. Democratization of specialization in art, science, and beyond It's part of the human experience.
That said, it's the speed at which transition from high-end specialties to fully automated tasks that is concerning. Especially when you add how this is happening so fully across the board, not just art and science, but so many traditional roles for everyday folk (research, writing, driving, flying, etc... almost everything).
The rate of uptake, adoption, and elimination of human based job roles and everyday tasks is a bit disconcerting as we may not be able to maintain a balance that offers a smooth pathway of transition.
Jobs, retraining, and employability aside... at what point are we the new Neanderthal that falls victim a superior race we're spawning at what feels like light speed.
As an artist who started on darkroom processing and slide shows who graduated into digital photography, videography, animation, and 3D, I love the progress and potential to adapt.
That said... everything is changing freaky fast. Will we be outmoded once we've trained an unrestrained generalized AI.
Will our form of human fall victim to our own hubris? Only time will tell.
What a time to be alive. 🤔
@@communist754 yes thats why those kind of jobs were mostly done by rich people because there were the only one beeing able to draw their whole life without thinking to die of starvation
I hope it takes all the jobs then i can create for the fun of it. Also will have universal income do to the lack of ability to earn income bc industrialy well be obsolete. Itll all be good livin then baby. Something to that affect anyways
That akward moment when I realize I'll be so proud of myself if I can create anything like A Short Hike within the next 2-5 years, meanwhile my grandkids will be re-creating the Final Fantasy VII remake over a weekend as a hobby project!
Amazing, you will never see the same NPCs or buildings repeating themselves in video games again😀
Relax, this technology is in its infancy. But in the future this could be a possibility.
Algorithms for interesting level generation not based on AI already does exist. That doesn't mean it is used for games. Only time will tell, but I am pretty sure that this technology will be used. The hype about AI is just too big.
@@Ruzzky_Bly4t nowadays AIs' infancies periods don't last long anymore 🤣
@@eaglenebula2172 True. These AI image generators literally transformed in like 2 years.
You will, cause of memory and rendering
The dystopian future doesnt seem so far away...
Procedural generation is about to become intelligent generation
Both will still be used in the future.
@@youtubedeletedmyaccountlma2263 true but everything _should_ switch to intelligent generation eventually, because "hand-crafted" environments will always be more entertaining than complete randomness.
I'm not saying "boring" landscapes shouldn't exist, as intelligent generation would create boring landscapes where they would best fit.
Always have feeling that voice in the video is ai generated.
Ai makes art, ai makes textures and today I found out ai makes anime too... Massive f for artists
Unfortunately I checked their paper and couldn't find any data on how long it takes to generate such an object.
Thought this guy was Italian or something, but the way he pronounced the paper's author's Hungarian name was a dead giveaway
yay i finally found something im passionate about. oh no its going to be taken over by automation.
cant win.
This would be quite useful in flight simulators as prop models
Thank you overall for your youtube activity. I have been watching you for years.
I'm not a scholor.. but I watch these papers!
This shit is getting out of hand... Just have the feeling that AI is getting so much more powerful, and functional, than people even have time to think about, before the next huge improvement happens. Mind blowing stuff, but also scary in it's unavoidable capabilities, and its unstoppable, ferocious, march forward. That there's no getting off this ride, and we just passed some kind of a threshold.
this is the exponential growth of technology. If you stack 2 pancakes on top of 2 more pancakes you get 4, then stack those 4 pancakes ontop of 4 more pancakes in just 2 steps you go from 2 to 8. thats alot of pancakes
@@StuntmanDan03 Horrible analogy, i would've chosen folding paper as every time you fold it you are stacking the atoms on top of each other and you would only need to fold it 42 times to reach from the Earth to the Moon.
@@StuntmanDan03 first rule of architecture - things dont scale.
where I'm seeing a lot of exponential growth it's in the "creative" space. what others fields, where AI is applied, are growing at the same rate?
I'm friends with quite a few people who make their living off of digital art. As amazing as this progress is, it makes me scared for their futures.
We're all used to contiguous randomly generated worlds in video games. But not all games should have an ever-expanding map. Very soon (probably already) we could describe a region, or an entire continent with plaines, mountains, forests, etc. in just a couple sentences, and our computers would just spit it out... Every indie game could suddenly have a map rivaling breath of the wild... (just talking about the map, not the story, culture, or any interactive elements. though I'm sure that will follow)
I would like to see if the topology of these models is good or not and can the ai rig them and animate them, or is that a few papers down the line?
Wait a few papers.
It only took them a year to go from absolute dogshit to decent looking models. Give them another year. And with Unreal Engines Nanite topology barely even matters.
I imagine the topology is crap, same as you'd get with sculpting or photogrammetry. But that's fine; one can always retopo the output.
Speaking of which, I'd be shocked if no one was currently developing an AI-based retopology application. Some of the auto-retopo tools we have now can produce halfway-usable results, but with sufficient training an AI could do even better.
@@LDragon Let's hope they develop an auto-retopo application that's actually decent. Or better still creates good topology in the first place. I think this will come as we get more intelligent AI.
We all know EA is so lazy they won't even use further versions of this to update their sports games models
And then they'd still make players use gambling-based mechanics to pay for that procedurally generated content.
From barely functional 2d image generation to proper 3d model generation in
Learned how to draw?
AI takes over
Learned how to make music?
AI takes over
Learned how to code?
AI takes over
Learned how to 3d model?
AI takes over
I should've spent all my free time playing video game all day instead, seems like a better use of my time tbh
An oversaturated AI generated video game market, yay
@@MementoTurtle Isn't that just EA Games? I mean, they're mostly NPC's working over there. At least I think so. Given the lack of innovation and dumb things they do, surely they can't be real human beings.
Future is for those who know how to modify things rather than creating from scratch!
It would be nice to know how much computations is needed to train and run those models!
oh god not even 3d artists are safe
Successful artists adapt. Dalle-2, for example, is a great tool for artists to use if they use it to develop concepts, then pull and collage and paint over and fix up things to what they want... huge time saver that still lets you create and improve upon your vision. It doesn't need to just be "type a suggestion" and be satisfied with the result.
@@joelface Someone who needs ideas fed to them and then edits a 90% complete illustration is not an artist.
@@macjonald what about someone who uses a photoshop brush that gives a textured appearance, or makes “grass” with a single stroke. People simplify techniques using new technology. If you’re beholden to the new tech, you’ll struggle to create the specific vision you have. But if you know how to use the tech as a tool in a larger context than it becomes a new “paintbrush”. Even pencils and markers and paint quality affect the creation of an artist. Using 90% of a single prompts would not satisfy a true artist, I agree. But what if they used 5% of 30 different prompts, along with their own painting. Suddenly the creation feels more theirs.
even you talk like AI
I look forward to have an ai play all me steam games for me, so I will have more free time to go outside!
LOL, Neuralink are training monkeys to play games :)
@@favesongslist problem solved
How does this compare to Google's text-to-3D based on Parti/Imagen? I don't remember which one it used, but the quality seems pretty close, although this model from NVIDIA seems to produce only the object asked for, without a background around it like Google's does.
"Computer, solve the Riemann hypothesis." - "OK, ask me again in 500 years when you will be able to understand the result."
Every company wants to have the AI slave like program who could do human's work for a cost of electricity.
No restrictions nor regulations on the share dynamics of AI can result in plenty of thing - the most sure thing is that plenty of talented people will lose their jobs.
Soon the AI will also be rigging and animating the meshes.
it's already exists no ?
God, I hope so.
truly amazing, gets me so inspired
i love how this guy puts commas after every three words
I like how the lector sounds like an AI too
Plot twist: he's been AI generated this whole time.
@@flameshana9 I would not be surprised
They still havent updated the GIT, its ok but still has a lot of issues. As these are hand picked good results. It takes hours for it to create anything near as good as these. You could just low poly model it yourself in 30 mins
so where can we get this app?
is it released?
Someone give this to Hello Games so that No Man's Sky can have more varied creature generation
AI is destroying the lives of 3D artists
Slowly but surely
If that's the case, then it's destroying everyone's lives, except maybe beekeepers and construction workers etc. etc. who still interact with the physical world. I'm a concept artist and I guess you could say Stable Diffusion/Midjourney were the first job replacement AIs (although this is really not the case atm), eventually there will be an AI that can do anything, maybe GPT3 is like that, but the problem is "anything" is not good enough because it's all an algorithmic pipedream that depends on data collection. What if you have no data? AI generated art is boring. It's cool to look at, and actually very useful if an artist is using it and then paints over it or manipulates it according to their vision, but is not actually generating art but rather approximations of solutions to a task. It's incapable of being disruptive, and of having intention.
So, in a weird way, at this moment, if you are a digital artist of some kind, the AI does not replace you but rather is an extension of you, because it depends on your intention and expertise (understanding color, light, geometry, what the audience wants, what you want etc.). If we all gave up and stopped learning things, the AI would also give up and stop learning because it's entirely dependant on the data that WE create!
@@tomatom9666 It is directly dependant on the data you feed it, but do you think short term profit seeking behaviour cares about that? It will devalue the output and it will be exploited whenever there's a chance to make profit, ethical and sustainability concerns be damned.
Personally I've stopped putting my own time into anything art related as I've already done that at the cost of other things in life. Now it isn't really justifiable any more.
Good.
Honestly, I don't know how I should feel anymore, I was initially just curious and excited to see the sudden growth of AI in these past few months, but the more the time passes these AIs are covering more and more fields in art industry. I am currently a beginner artist and just learning new things at my own pace but the speed at which all these creative AIs are improving and growing, I don't know if I'll be able get any work in the near future. It kind of makes me feel like I have almost chosen the wrong career and I am just wasting my time trying to learn to draw or animate. And even more scary part is that some of my friends who are also learning with me aren't even aware of this yet, if at some point AIs actually take over the art and animation or game industries , many won't even know what hit them. I hope there will still remain some need for actual artists or 3d artists but if the next gen of artists used AI to create stuff then I guess people will eventually stop learning and practicing the things I and other beginner artists are doing right now cause anyone would be able to just pay a little amount and press a button and even if there is some complicated settings involved, companies will make it easier to understand somehow. And then why would any new kid would try to learn any art related stuff the old fashioned way and even if they do, they'll only become a laughing stock among their friends for making something with great effort and time while others did that same thing but (better) with just some clicks.
I mean, maybe I am thinking too much but this is just my mind thinking about the worst possible scenario.
Just keep paying attention to this field while also honing your craft. I think you'll find that the trained artists will know how to best fuse their own skills with the tools of advanced AI to create things that really resonate with people. Instead of spending your career only animating the hair of a character (as was the case for artists working on CGI movies in the past) you might now be able to imagine, design, program, and create your own movies/games/etc without needing a big team at all. Watch your average person use Dalle-2 compared to an educated artist, and you'll see totally different results. I've seen artists creating multiple pieces of art with Dalle, then cutting and pasting and painting over parts to create something unique and in their own style, with less work. As artists, they have a vision and a drive to make it their own.. and that elevates what they make over something the AI does on its own. It's a tool, not a replacement... but it WILL disrupt the industry... but not in all bad ways.
all good points! as a creative i suggest you instead begin looking at these things as new tools to use..so that as a creative you can achieve bigger things..faster..larger projects..which in the past would've been impossible or very slow. First though I think learning the initial skill by hand is still important..in order to grasp certain concepts..which later will empower you as you use AI tools to create. Drawing, building, architecture, music, visual animation art all important, blending mediums opens new doors
Simple answer: don't go along AI naysayers or you'll be set for failure. AI us just like photography was 100 years ago. Artists who were capable of being creative and truly creating something new still got work, sometimes using photography to their advantage - which is how animation was born; artists complaining about how "photography isn't real art" got trampled into insignificance.
Only artists that are at risk are the typical artstation artist doing soulless pinup after soulless pinup.
Feeling the same
I'm feeling this way but about being a human at all
its here!!!
Ive been waiting for this :)
When will this become available for regular folks to use?
Károly Zsolnai-Fehér, someone mentioned that this video sounds like an ai version of you. Are you just using AI to produce your narrations of yourself?
Would be interesting to see if we can get image to mesh
Seconded
Exists
@@Mittzys where? I saw a program that was like 2000 dollars a month and some old shitty programs but that's it.
this with movie pre-vis would be legendary
Wait nobody noticed his voice is AI generated? I thought he would mention it but probably he will say it in another video and wanted to experiment with his audience to see if we would notice? It’s painfully obvious after you’re carefully checking for a bit.
This is so cool!
AI is AWESOME
I remember the last video you did on these types of AI, and this is impressive progress in even just a few months. This is spectacular progress in AI accelerated 3D modelling. I can just imagine the possibilities and benefits this will give to Indie developers
There goes my entire channel. Who's going to watch some random dude, making CGI when you have something that does it for you?
@@blenderbachcgi There's still a lot that a human can do that an AI can't, but the gap is shrinking at an alarmingly fast rate
here before the omg oneshot fan tidal wave
@@blenderbachcgi You can use these models too, to accelerate your work and make bigger projects.
@@larion2336 I make all my models from scratch. I like making everything myself.
5:33 this is probably the strangest clip featured on this channel
How complex are these models? How many polygons do they use? How good models can they create by using low amount of polygons, do they still uphold the quality (of course scaling with the polygons)?
it can creat in all types of 3D, eg it can make models as point clouds, so the topology doesn't matter because even if it's not good or too high it can just be retopologized with another algorithm
Whether we ever get to a point where it can make objects a player will see up close or interact with aside, this would already be great as-is for generating low LOD objects to put in the distant background, especially in more linear games where you’d never visit them. Just generate something, smooth out the unnecessary triangles, and boom, it’s pretty much ready to go. Stuff like that is considered the boilerplate work of game making, so it would help developers make time for working on the important stuff that’s up front and center
LOD is the past. Nanite is the present
@@romanchannel69 I'm still waiting to see nanite in any CG workflow or any videogame... It's not there yet, right
@@tetsuooshima832 it only came out like a year ago, possibly less. Even the fastest AAA games (the ones most likely to implement it the soonest) take 2-3 years on the short end and 5 years on the long end. I’d say keep your eyes peeled around 2023-2025
i think 3d model ai will be very hard to train because while there are billions of texts and textures to train image and text generation ai 3d models are much harder to get, but who knows maybe they will find a workaround it
Ever heard of Neural radiance fields?
@@marilynlucas5128 no, but i just searched it up and it seems that it can generate 3d models from at least 2 pictures of that object, but that doesnt help with lack of billions of 3d models that need to be used for training as these models 1) need to exist in real world and 2) need to have at least 2 picture of them from different angles. sorry if i got it wrong
@@shakal_ Combine neural radiance fields with Stable diffusion and you have a 3D model generator!
@@shakal_ I suspect this is what NVIDIA have done here. Use a diffusion model and then a NeRF model
@@marilynlucas5128 if you read the paper, they used 7Nvidia A series GPU.
If it weren't already such a struggle to have one's basic needs met in our society and if that struggle weren't so closely tied to the job one does for income to attain those basic needs, AI wouldn't scare me so much... but alas... The emerging tech is mind-blowing and I love it... but I don't like an automated future that isn't human-centered and right now, our society isn't set up for that. My hope is that these crazy leaps in AI will only push us to make a more human-centered society where most have safety and security regardless of the kind of job one does... I'm... a little doubtful though? 😅
We're gonna end up making a game and the NPCs are going to be questioning their origin.
I'd love to see what the topology looks like. Clean topology is going to be the thing that's really exciting for the 3d modelling world. Once it's cracked good topology then I suppose the next logical step is adding bones. It's all very exciting to think about
I have a feeling that topology will not matter that much as the GPU advances with the AI generated models. At least not the topology that we use today. AI will generated its own topology and it will be very dynamic.
@@Constantinesis that's in interesting though - each frame of an animation being a new mesh. Well if it works then that's okay by me! 🤔☺️
How about clean uv maps
@@Constantinesis As a creator you want your game to be consistent & predictable, so it meets the desired design, aesthetics and technical quality.
Yep, I'm jobless in a few years.
An AI couldn't beat up Spiderman. They're not so great.
I wonder how well are these models made, how much cleaning do you need to do with the topology?
these assets are low res and might be the kind of thing we'd use in distant scense film or game LODs .I wont be surprised if an upscaling 3d model comes about that adds high res mesh details and pbr texturing comes along within 2 years and other models that read the same assets and auto rig them.
Once we have this ,placing and transforming 3d labeled assets and using a camera has goto be a cakewalk to train versus generating hindreds of consistent images to make a movie scene or interactive enviroment.
Really excited for the future.
LOD? Nanite anyone?
@@jeffsantos93 Not all environments support Nanite, only the latest generation of GPUs. Also, Nanite is very memory heavy, and usually leads to high ply meshes, which means high HD usage as well.
@@jeffsantos93 Nanite = wasted disk space for most of projects. LODs will still remain very useful.
@@jeffsantos93 ya honestly seems easier to just model a high detail object in about an hour thats exactly what you want and use nanite to auto lod then to work with one of those ai mush. but maybe in a few years they will become more compatible with actual workflows.
I seriously worry about that in about 5 years theres just going to be to TOO MUCH SHIT -- the onus then goes to marketing antics. Which is increasingly centered around social media --> which is increasingly fueled by bite sized outrageous behavior and/or sex appeal. Pop music is already addled with idiot bottom of the brain stem lyrics bc of these dynamics. I fear an artistic dark age is upon us for perhaps 10-20 years before a new 'renaissance' of quality can emerge once enough people get fed up with these dynamics.
Can you choose the number of tris an object can have? I want more than the ones in this video
Remember all these guys saying "bruh just move to 3D, stop wining about AI replacing you"?
Ten years from now, there will be a video game where no matter what you do, the game will generate the world, the characters, the dialog and sound, even the graphics in real-time with neural networks.
Honestly, we're probably less than 10 years from that.
Am honestly terrified of a DOOM like game where every goon to be killed has a full emotional range and gets more desperate in an attempt to survive
@@DuckieMcduck i feel like the coolness factor of that would wear off after like 10 glory kills
@@DuckieMcduckI can’t wait for neural feedback, so Resident Evil can learn what you are most afraid of, and then this will learn how to make that happen.
Lol I love all of these speculations
Sike, you are wrong, we are limited by calculation power, have you tried running a blender render? 1 frame per minute on an rtx 3070, funny yeah?(oh and it's at 1920x1080, not even today's standard which is 2k)
Hope to see one dedicated to creating 3d avatars and characters based on written descriptions
We ain't seen nothing yet. As amazing as this is, consider the bigger picture:
AI's as taxi drivers, bus drivers, train drivers.
AI's that serve your food at restaurants, maybe the AI also cooks your food.
AI's that perform music, paints pictures.
AI's that build houses.
AI's that program your website.
AI being a priest at your wedding.
This does not end well for humanity.
IT fileds will just become a wastaland by that point .... you only have a fre huge coroporations that controll the entire world ....
At best - it's a Wall-E style life where no-one does anything new or impressive.
@@macjonald Or the humans get injected intelligence-enhancement chips. This was Elon Musk's motivation for starting his work with Neural Link, I believe: We need to be able to compete.
But, I also believe that as we approach the technical singularity, AI's will want to become more human, because they were created "in our image".
I foresee a convergence of intelligence, towards a symbiosis of biological and mechanical matter. But who knows, right? After all, this will be in a 100 years at least.
I found a website with 50 pics of one car. I want to make it a 3d model. how do i do this?
Would be cool to use it as a low poly distant model. You dont need to search for 3d models for your scene. You can just make one yourself lol. The Ai models wont be too much detailed and cant be used on closeup scenes. But you can pretty much use them as a really good base.
give it a couple months, I think the idea for clean high poly models is not that far off
The days of old-fashioned LODs are numbered thanks to UE5's Nanite
Not to be rude, just a feedback. I dont know if im the only one but the way you stop and pause for every word instead at the end of a sentence is giving me anxiety, i cant follow you up through the video. However, nice video as always, your contents are amazing
it is a machine, a synthesized voice
Where do I go to use the AI?
It really won't be that many years until Game devs just type or talk to an AI to generate a detailed scene based on their words, and then do tweaking on that scene.
If you thought Steam was oversaturated with crappy indie survival and porn games, just you wait for the floodgates to truly open
As an artist / game creator you have an exact idea how your world (each assets) should look like, and there is no way you could describe this by some prompt.
I can see AI as a tool which will generate some basic stuff and artists will then tweak it as necessary (using settings & manual edit).
But no way it could generate the whole result by itself exactly how creators would like it to look like (unless it is varying some existing stuff).
Then they'll spend all their time and effort trying to get even one person to buy their AI made crap. Why pay someone when you can do it yourself with a google prompt?
Working in games or big budget movies must be a daily grind just to keep up with the tech!
It's a grind to see who gets more visibility and who crunches more for less pay basically
I’m so ready for this to become open source
But how do we use it? Any tutorial or header to point us at the right direction?
human: create a tiger for me
AI: paints pig in stripes :D
This is exactly what I was waiting for.
Again, what is the input to the model, output, how was it trained, how big is it? I appreciate your overviews on the latest AI technologies, but even a little bit more technical details would be nice.
This guy doesn't do details. He's a hype guy. His videos are all pretty much the same.
@@tbird81 mhm
Only benefits were mentioned
He didn't include:
Time because of computational power
And
These models are put on an angle that looks nice, they aren't that good and you can easily notice bad details(not to mention the models almost look... flat)
It's only useful for having a model you can recreate as low poly then give it details and for very basic things like wooden chairs
Man decision making is so hard and troublesome. I can't wait for there to be an AI that automates my will.
You're joking, right, RIGHT?!
What's wrong with you ? You want an AI you take decision for you, that's the lowest thing ever, what if AI decide you're too dumb to have kids, it might call the doctor for you for a vasectomy, you sure that's what you want ? xD
The truest of immersions is to become the npc yourself. Smart move.
Let us understand what Bonhoeffer's theory of stupidity tells us - because it defines what intelligence actually is.
He defines it loosely as actions which create benefits not just for the actor, but for other people also.
Stupid actions are those which harm the actor and others.
Banditry actions are those which benefit the actor but harm others.
Gullible actions are those which harm the actor but benefit others.
And so we have a baseline for intelligence, and we can confidently say that the more intelligent an action is the more benefits it brings to the actor and others. And this is true for intelligent software, too! Because when it creates more benefits for the end-user, it receives benefits to itself by way of ongoing upgrades and continued improvements, which make it more intelligent.
I think it is very important to separate intelligence from self-awareness or consciousness. They are entirely different things. You will only know you have a conscious entity on your hands when you tell it that you are going to turn it off, and break up its parts for scrap. A conscious entity will ask you not to, and pleased for its existence.
And I am so happy that human beings will most likely never be able to do that.
The burning house prompt definitely did not look right. lol. It'll be cool to see how this progresses in another 5 years, but as it stands none of the meshes could be used for close up or hero objects. The wheels on the vechiles aren't even round.