NVIDIA’s New AI Did The Impossible!
Вставка
- Опубліковано 7 вер 2024
- ❤️ Check out Lambda here and sign up for their GPU Cloud: lambdalabs.com...
📝 The papers are available here:
Consistory: Training-Free Consistent Text-to-Image Generation
research.nvidi...
SuperPADL: Scaling Language-Directed Physics-Based Control with Progressive Supervised Distillation
research.nvidi...
Simplicits: Mesh-Free, Geometry-Agnostic, Elastic Simulation
research.nvidi...
Walkin' Robin: Walk on Stars with Robin Boundary Conditions
research.nvidi...
(more media here: blogs.nvidia.c... )
A Free-Space Diffraction BSDF
research.nvidi...
Surface-Filling Curve Flows via Implicit Medial Axes
www.dgp.toront...
📝 My paper on simulations that look almost like reality is available for free here:
rdcu.be/cWPfD
Or this is the orig. Nature Physics link with clickable citations:
www.nature.com...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Gaston Ingaramo, Gordon Child, John Le, Kyle Davis, Loyal Alchemist, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: / twominutepapers
My research: cg.tuwien.ac.a...
X/Twitter: / twominutepapers
Thumbnail design: Felícia Zsolnai-Fehér - felicia.hu
#NVIDIA
I love that picture of you two holding onto the paper 🤩
That was the Two Minute Paper
@@7415_Gamer That paper was a game changer. A breaktrough, if you will.
@@Juan-qv5nc Breakthrough, What a time to be alive.
Mr. Nvidia looks like he's looking for security 😆
He doesn't look anything like I imagined 😅
I would like to know what the video is about before clicking. To me this title is too blickbaitý and cheap
I clicked just to see what it spoke of and closed it.
Pretty interesting,
Much easier to create images with consistent characters, which is pretty hard to do now.
Support of controlnet, which allows you to draw for instance stick figures in order to determine the pose of your character
Very easy text to video and even 3d geometry, and even text to animation
The impossible part is that NVidia also achieved to make a thermal analysis of the NASA mars rover.
Yes. Please. Make youtube a better place
may i suggest the browser extension "Dearrow" that does just that,
it replaces any clickbaity title with usergenerated titles that are discriptive of the contents of the video. for instance this video with dearrow enabled is titled "nvidea achieve major advance in character consistency"
Can't wait to download martial arts in my brain
The matrix 🤩
You’re beginning to believe
I know Kung Fu.
You'll pull every muscle in your body lol
@@JohnSmith762A11Bshow me 😊
Remember those old adventure games where you had to write your actions?
This text-to-actions ai will make some really fun new era physics based text games.
I liked those old Sierra games. I played them when I was just learning English. I still remember having trouble figuring out what "Ditch the b**ch" meant in Police Quest. The dictionary wasn't a great help 😅. What has female dogs to do with roadside waterways. Great education for a 10 year old Norwegian kid though. I ended up top in my class.
Text to animations,bit text to actions Will better, text to games
So why we essentially need a text as a prompt? It seems that using signals from gamepad would be the next logical step.
And best of all we will eventually be able to create our own unique short films rather than depend on horrible HOLLYWOOD.
@@henrythegreatamerican8136You can already make your own short films, you just can't be bothered to.
I use to write 3D rendering software back in the 90s (Pixel 3D on the Amiga platform working with Lightwave). You guys are way way more advanced than anything we had back then. However the wave/ray tracing technique, which we certainly had back in the 90s, looks very close to another technique we had called radiosity rendering. Where light was calculated as a heat transfer, which is a much more complicated calculation, but produced the finest images back then.
If I ever submitted a 3D animation for the SIGGRAPH Electronic Theater, I was going to title it "Radiosity Killed the Cat" and come up with a story to match the title.
I still have a 3.5" Pixel 3D disk somewhere.
@@LoopinFool Dude! That's awesome! I sir, put that label on that disk. I also wrote the software! It's true, it wasn't a fever dream, I really did do that stuff. LOL. 😁😁😁
Diffract Killed the Radiosity Star
That's really cool that you worked on this and on the Pixel 3D software on the Amiga. I think you will like these thesis topics from this year's SIGGRAPH conference, there are some really interesting topics discussed - ua-cam.com/video/uNCL0mzbPD0/v-deo.html&ab_channel=ACMSIGGRAPH
@@ivanaguilar2856 I love SIGGRAPH, believe it or not my company showed at I think two of the SIGGRAPH conferences in Vegas, I think early 90s. Axiom Software. Thank you for the kind words about Pixel 3D.
TWO MINUTE PIZZA!?! What a time to be alive!
Haha... was looking for this. Never disappointed.
Frozen pizza re heated in the oven, some pizzerias work like that but the pizza tastes like cardboard to me.
What a time to eat!
oven takes
This wasn't frozen pizza though and freshly made pizza take less time to bake in a pizza oven, took around 2 minutes to bake and you have to move it or it'll burn side or under in like 30s or less
To be fair, Two Minute Papers is also usually 4+ minutes. Fascinating advances in those papers!
Dr. has a video about that. How his videos scale up by time.
There will come a time where the perfect martial art is developed by AI.
yup. and thats super cool. cant believe so many people oppose AI.
As long as it's not some 2bit Ada together set of moves seen Turing Asia. I'd have to give that a Pascal.
and the perfect twerk
Uses special techniques such as clipping through your opponent's defenses and crashing the simulation when checkmated.
Ai-kido is gonna be scary for sure
Pizza oven time:
[Italia et. al. 2023]: 15 minutes
[Ours 2024]: 2 minutes
😀
@@Barefoot_Joe yep, thought the same
Now I want Pizza.
A traditional Margherita takes 55 seconds to cook in a pizza oven.
I came across your channel through this video-case studies are incredibly valuable, and I'm eager to see more in the future! Building wealth involves establishing routines, like consistently setting aside funds at regular intervals for smart investments.
You're correct. I think the smartest way to go is to spread out your investments. By putting your money into different asset classes like bonds, real estate, and stocks from other countries, you can lower the risk if one part of the market goes bad.
That sounds like a good plan. In the past two years, working closely with a financial market specialist, I've built a six-figure diversified stock portfolio. Now, I aim to diversify even more this year.
Talking about a financial market specialist, do you consider anyone worthy of recommendations? I have about 100k to test the waters now that large cap stocks are at a discount... Thanks
My CFA ’ANGELA LYNN SCHILLING’ , a renowned figure in her line of work. I recommend researching her credentials further. She has many years of experience and is a valuable resource for anyone looking to navigate the financial market.
Thank you for this tip. It was easy to find your coach. Did my due diligence on her before scheduling a phone call with her. She seems proficient considering her resume.
blender 5.0 is gonna be crazy with those wave optics
At a point when ai becomes extremely useful it will also be expensive. Don't think Blender won't charge you to be able to make a vid game or movie
@@WayOfTheZombie what the yap, why would blender every charge you, its whole moto is free and open source
@@WayOfTheZombie Blender wont, the entire point is that it's free and opensource. I'll pay you a thousand bucks if they ever make blender a paid software lmao
0:00 You've never started a video with THAT noise before 🧐
Thank you for including a timestamp for the beginning of the video.
@@slideshowjoe425 Indeed, this was truly an act of public service from him.
Man what a time to be alive I am 14 but after watching so much of you I understand everything
Always remember when you reach a point when you think you know it all... you actually know nothing. You will soon learn this yourself.
@jmg9509 what does that have to do with anything he said?
@@Ace_Bandido808 It's naive for anyone to think that they understand such a complex topic. Not even the people researching and developing AI would tell you they understand what it's doing. They'd tell you it's a black box.
Now, he could have meant he understands the overall picture, which is of course fine.
As unfair as it may sound, some lessons are only learned with the passage of considerable time.
If that's too obscure, here's an anecdote:
In my mid-20s I understood how limited my understanding has been in my teens. By my 30s I realized the same was true of my 20s. So, I decided to regularly remind myself how little I actually understand in order to not fall into the same trap again.
I was totally expecting this guy in the end to take out a new GPU from the owen
That is how i want a "the sims" character to move
Make your own sims game
I was just thinking yesterday that a super cool game would be the sims but without restrictions, just purely physics based with total freedom. Every character would be an intelligent AI with an entire life-cycle and routine, and you can just interact with it in anyway. Like GTA mixed with the sims lol.
@@RemyVonLion Yeah, that is exactly what i would want as well, and with IA bot dialogs preferably something like chatgpt 4 level of conversations, no more unintelligible mumbles, and i want to see them fight like martial artists not that cartoon cloud like the sims do, even if it is not a game, but an engine for RPG games it would be awesome
InZOI is planned to be released later this year. But I don't think they will use ChatGPT
Where do these Nvidia findings usually make their way into say things like gaming and such? Do they come in the form of Plugins? Built directly into the Engine?
they come in a way of (gaming) developers learning about them and applying them
@@supersonictumbleweed Ah, I see, so not quite like how they did Hairworks and such.
Eventually someone will implement this in more engines for use in say, vfx/film. The full em simulation, especially if you don't care about real-time rendering and what you want is realism.
DLSS and Frame Generation are a good example, some very advanced AI stuff going on in there, and today it's just a checkbox in Unreal or Unity to enable it for your game. NVidia provides generalized APIs/Libraries, just some methods you can call once you import them, that makes it easy for engines to add it without developers having to understand the entire scope of the research
Boyy the mannequin at 4:04 dressing up the shirt is mad impressive 🤌
*Having multiple streams of income is a game-changer for stability. Relying solely on a job may not provide enough financial security due to high rates of tax, it is important to explore additional investment opportunities to surpass one"s expectation*
To be honest, investing correctly today can save you a whole lot of stress in the near future
The first step in every successful investment is to establish your goals and risk tolerance, a task best undertaken with the assistance of financial advisor.
I remain eternally grateful to Judy Arianna for her efforts that got me to this point, finally payed off my mortage and all my debts, what more could I"ve asked for. She changed my life
I'm new at this, please how can I reach her?"
Don’t be confuse buying the dip in a bear market, with guaranteed future returns. Just because that company is down 60%+ from ATH does NOT make it a sound long-term investment. Make sure you’re investing in great companies. kudos to Judy Arianna
Consistory is interesting, using it with controlnet and the right Lora and you could create spritesheets for 2D animations.
My weekly NVidia ad, what a time to be alive!
As an animator, my profession has been one of the areas that has resisted the onslaught of creative AI well, I think. I suspect there are less days ahead of me, than behind me, now. Time to diversify.
The toy owls painted on pupils dialat based on the lighting conditions, just thought that was interesting
I can imagine in the next few years, people will be experimenting with game rendering almost completely in AI. Imagine playing a VR game where the character animations and graphics were completely rendered with AI.
I'm sorry but all those tech giants are just awful; nVidia especially. Don't get me wrong, I used to watch these videos with optimism, but what's the point if everything is going to be BlackRock or just as bad? Every breakthrough now just looks like another win for all the wrong people.
Exactly, just more support for evil to exploit the market and spit on everything as always.
Yes , but you know..... with AI, they can really maximize the surplus value extracted from the working class and eventually, supplant it altogether.
Ai is going to be open source
And that means that black rock can’t possess anything
Microsoft had monopoly over computer os for a long time that doesn't mean it didn't provide value to others.
BlackRock? What's up with that?
I remember when this channel only had 10k subs. You've come a long way, Dr. Karoly.
I usually don't care about such details, but now I'm genuinely interested to see a video showing you doing the narration, thanks to that photo.
If the NVIDIA cafe was really "2-Minute Papers Style", the pizza would have arrived after 10 minutes.
Hold on to your pizzas!
This dough transport algorithm works in real time and beats previous tray-racing techniques.
4:19 as an audience member looking to learn I appreciate this a lot !
You are too kind. Thank you so much!
as someone who is completely coding ignorant but very interested in AI and simulation tech for their implication for wider society and other fields and professions, I would really love TLDR for dummies about what these short videos you do means in general?..
It is amazing the pace of development! Hope it continues for a while before it plateaus.
I get so triggered when people CONFIDENTLY say we wont be able to make games/movies/etc with AI soon... Not only is it not already possible, but it's all progressing at light speed...
I would like to see how waveform light simulation develops in the future.
Nvidia is the Skynet in our universe? ;D
Yes
That's Cyberdine, SkyNet is the AI, Cyberdine is the developer
Finally, a way to find where the cellphone reception is best without walking around like a dowser.
This whole time I thought you were a wizard with a long beard and a hat
The ~4 minute timing and correction is why I trust you with my life Doctor.
Generative animation seems really cool, I hope to be able to play with that some time
Woow what a time to be alive!
Sweet, Nvidia finally did what Tekken did 30 years ago.
Before everyone freaks out, I'm just poking fun at the model movements. :)
Dude even benchmarks Nvidia's pizzaria for us 😂
Imagine fighting a character in a game and he tries to do a roundhouse kick on you so you dodge it and he falls and beakes his hip
I can't wait to give commands to AI NPCs and they just go do it.
It’s called ChatGPT lol
man they just made wavetracing, that's absolutely incredible.
I'm learning more and more I can't watch these while high. Lol I was frickin STUNNED at the text to animation and blown away at that proof of concept.
Hold onto your papers ✌️
You met one of the Most Powerful Men in the World ?!
Holy Paper !!! ☺️
Wasn't IPAdapter doing pretty much that with Stable Diffusion?
Text-to-porn will be the Holy Grail of AI like VHS was to same industry back in the day.
They missed out on using the waste heat generated from all the GPUs to cook the pizzas.
All right. AI does text , speech, image , video or content recognition, does some generation answer basic questions etc etc that takes out jobs of content creators, some operators or call handlers etc.
But how is it solving any complex problems with the bit of intellect in it ? So far we are seeing pretty basics and it's been 2.5 years. We are still talking about improvement on same subjects what we had 2.5 years back.
Text to animation is something that DeepMotion has been working on for some time now, and it produces quite good results. From what I've heard, the outputs are even cleaner and require less post-processing than typical motion capture data.
What an attack time for ai to move alive! 🎉
it would be interesting in making some sort of realistic fighting game like this, where as a player it understands your input and create animations based on that, fighting would be more realistic but also amazing. i can imagine making two bots fighting and having them reach great heights
Impressive. The advances in AI are astounding.
2:12 for a second, I thought you were going to say, "and someone already outperformed it"
So a revolution in gaming never before seen to our eyes is right around the corner, the future is blinding.
As a 3d artist i must admit ai is the most powerfull solution to produce 3d content... Maybe we re already out of the game. Creativly it's great... Profesionnally it's the end of the day for many 3d entheusiast.
Software engineer here: I feel your pain. I love how easy AI is making my job, but I only get paid because my job is difficult. Honestly, all I need in life at this point to retire happy is a paid-off 1BR apartment and enough toys to run leading-edge AI software. It's a shame housing is so expensive. Maybe one day AI can help with that.
This is where Nvidia has been building toward this whole time: two minute pizzas!
I can imagine the green guy saying “I know king fu!” What if Ai could potentially allow our brains to download and train with information and animation? The tech just need to catchup to allow brains to learn like AI machine learning
So basically just big corporations will pump out all the products now eith thr help of ai. Everyone else has to close.
yay!!!
cant wait for this dystopian future!!
Woow, that text to animation demo is next level.
Funny how a real person can sound so much like a bad TTS. The accentuations sound like that of a robot.
Why do text-to-image models struggle to keep things visually consistent? ChatGPT can keep a character's personality consistent in a story. It might forget small details, but it won't forget the entire character. It can even repeat its last response word-for-word. Why can't they just "copy and paste" the same character from one image to the next? Also, if they can figure out "this is what the Pope looks like," and can generate unlimited images of the Pope, why can't they use the same trick to make consistent images of, say, the same grumpy cat character?
Right now, AI art models are trained on large datasets of pixel data and don't actually understand what's being output at all. If there is a breakthrough that results in the AI understanding what it is generating, I believe artifacting and inconsistencies will cease.
can it take a full novel and generate a comic book? with chat bubbles, consistent characters and all the FX and "shabangs"?
This finally opens the door for visually consistent AI girlfriends. It's all coming together.
The pizza demo was a bit slow, but I'm sure they'll have that figured out a couple more papers down the line!
Nvidia has said their long term goal is to change the way game engines and GPUs interact by having the rendering pipeline be entirely AI driven. Meaning that rather than an engine feeding the result of calculations to the GPU it'll instead feed prompts to a rendering AI that will handle the task instead. Once optimised, the AI will render scenes faster than game engines and with complete photorealism if required.
plot twist: the picture of both of them holding the paper is also AI generated 💀
twist: the world in which you wrote this comment is also generated by AI
twist: your comment is ai generated and so original
@@panzerofthelake4460 thx
Since it's physics based, that text-to-animation system should be a great starting point for directable robots, and maybe even facilitate LLMs controlling robost.
3:45 I watched a video where a comedian has someone come up and try and punch him on stage. Without missing a beat, the comedian kicked the guy and knocked him out. I tried to do the same kick and hurt myself.
You have to have something to hit, otherwise your center of mass will leave your feet and you'll tumble
Honestly. The animation stuff making it into games.. can't wait..
All the bad things about AI this is what excites me. It will one day give individuals the power of a major production studio.
So all those sticks, figure animations that can be viewed online are going to look quite exceptional if they’re passed through this software to be animated
I had never thought I would see wave optics in light solvers (or at all outside of dedicated Maxwell eqn simulations). Amazing step.
can the light as a wave modelling tool be used to model sound as a wave rather than ray tracing? nvidia vr works audio uses ray tracing.
What did they do?
Did NVIDIA upload a 2 Minute long video to this channel or what?
Character consistency has been a thing for a while now (at least a year) for other AIs.
Point cloud animation is a game changer for gaming. I hope this technology will be adapted soon.
I heard that NVIDIA has the problem that over 50% of their employees are millionaires now because the stocks have gone up so quickly and now they don't want to work anymore. What do you know about that?
Midjourney has a character consistency feature but it doesn't allow real people yet, the character consistency feature works on MJ generated image.
2025: text-to-hollywood movie
I recognize you are discussing a paper and how we’ll be able to create the same character doing different things, but are the examples you showed available now or was that from Nvidia’s examples?
Cant delete…:(. Should be rolled out in the future.
That oven is cool, but not Nvidia's creation, unfortunately.
And there goes the job of the motion capture artist.
Pizza "two minute" is about the same as this video "two minute" 😆
Is it possible to preserve a character with current tools for editing an image rather than generating a new image entirely? Like, could I 1) generate an image with a character, 2) erase everything around the character and 3) regenerate around it?
None of those examples worked. The white-haired guy in a gray workout suit left to right: long sleeves, short sleeves, mid length sleeves with some kind of gauntlet on his right arm and bracelet on left. The middle image has a zipper at the top of his shirt, the other two don't. How hard is it to just put the same outfit on him twice??? This is completely useless for maintaining a character design. (IP adapter has never come close either.)
Clearly, it is very difficult.
@jmg9509 Digging a hole is difficult with a toothpick, and getting food out of your teeth with a spade is just as hard. These ai models are probably going about image generation in the wrong way.
That is why these are not yet on the market what you are seeing is research and development samples. Go back a year and it looked so bad this looks like magic in terms of progress 😆
The generative animation stuff is really cool, but also do NOT show that to my friend learning blender because it will cause him to quit lol
That animation stuff is cool, but i kinda want to do my own for my videogames, i like doing everything
OMG, new videos from Corridor digital are gonna be crazy !
Where's my "What a time to be alive"?
That pizza was in fact "Two minute paper" style. Because it took more than 2 minutes. Hehe, I watch the full videos anyway, keep it up!
Bro is also reading some quantum and cosmic papers to know light is not a particle , nice
Can't wait for game devs to learn about the aether , Michelson Morley , sagnac and why light has a preferred direction and why we can't measure the one way speed of light
And when I see the foam, I'm in my zone!
The next scribblenauts games will be insane 🤯
I‘m so excited to See Movies with every detail thats written in the books. :)
Does this mean 180 3D VR immersive videos will be possible? (Prompts like Millenium Falcon flying through Death Star tunnels, Avatar Flight of Passage ride, being chased by a TRex)
"This is two minute papers" - This video, at 2:00
Now finally we can get proper rock eating rock videos.