NVIDIA’s AI Puts You In a Video Game 75,000x Faster!
Переглядів 145,388
27 днів тому❤️ Check out Microsoft Azure AI and try it out for free:
azure.microsoft.com/en-us/sol...
📝 The paper "Live 3D Portrait: Real-Time Radiance Fields for Single-Image Portrait View Synthesis " is available here:
research.nvidia.com/labs/nxp/...
Fully Connected conference: wandb.ai/site/resources/event...
MKBHD, iJustine, Brian Tong persona source: • Using Apple Vision Pro...
📝 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/articles/s4156...
🙏 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, 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
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Károly Zsolnai-Fehér's research works: cg.tuwien.ac.at/~zsolnai/
Twitter: / twominutepapers
@user-jc3nz7eb6t 25 днів тому
Every time I hear you say "What a time to be alive", I smile, because it actually is an awesome time, full of technological wonders!
@DuckieMcduck 25 днів тому
what is the wonder? the only thing this is good for is misinformation and scamming old people
@ender0998 25 днів тому
Also a time where our governments are fueling a genocidal and war criminal state ...
@btm1 25 днів тому
Meh, tech is not inherently good or bad, it depends how it is used
@SaumonDuLundi 25 днів тому
So what a time to be alive!
@ollydix 25 днів тому
I think, oh no my job is lost.
@soccerkenshin 25 днів тому
Wow, didn't know rtx 4090 is considered a "commodity" 😂
@M4rt1nX 25 днів тому
Well, within this field, it is
@southcoastinventors6583 25 днів тому
Maybe he already has 5090
@Nulley0 25 днів тому
Nvidia X100 costs a datacenter I'll woosh myself
@jooei2810 25 днів тому
I remember the days that you could get a top-of-the line graphics card under 1000 euro.
@vaendryl 25 днів тому
compared to data center grade hardware, yes it is.
@leemonguy 25 днів тому
I too love putting myself in a video game 75,000 faster
@ulogy 25 днів тому
Papers: Lost, forgot to hold onto them.
@moroteseoinage 25 днів тому
Remember to “Save as,” sirs.
@blackowl641 10 днів тому
He doesn't do papers anymore he does nvidia promotion videos. Haven't seen a real research paper in like over a year. UKposts money eventually gets to every channel, then they stop being about the same thing
@alotino 25 днів тому
Avatars looked horrifying before this paper.
@shadowdragon3521 25 днів тому
NVIDIA’s AI allows anyone to impersonate you on the internet 75,000x faster!
@skyefreeman9987 25 днів тому
Finally my persona can reach its full potential!
@rrggss5057 24 дні тому
Fun fact: input image at 0:54 is Cristina Cifuentes, expresident of the Community of Madrid in Spain from 2015-2018
@randomdude2086 25 днів тому
Cant wait for two papers down the line when it will be possible to make a full 360 from one image
@jooei2810 25 днів тому
Well it would have to make up data from thin air to achieve this.
@herp_derpingson 25 днів тому
Its already there
@randomdude2086 25 днів тому
@@jooei2810 Isn't it already doing that?
@randomdude2086 25 днів тому
@@herp_derpingson I guess now we have progress faster than videos
@hiddendrifts 25 днів тому
@@randomdude2086 >already doing that< technically, perhaps. but they're saying that, unless you use a full body pic, the ai would have to make up what the rest of your body looks like
@thicccrusade2302 25 днів тому
Scans your face and then you are in the database.
@SlabtheKiller89 23 дні тому
Once there is one (1) picture of your face on the internet, you cannot rule out someone will be able to make a 3-D deepfake of you
@takeuchi5760 22 дні тому
You say this like your image is not in some database already.
@SlabtheKiller89 22 дні тому
@@takeuchi5760 never denied that. Just think about Linkedin profile pictures that you put yourself!
@deondrex18k 22 дні тому
Can't wait to see my Doppler ganger@@takeuchi5760
@spongebobsqaurepants 19 днів тому
Do you have a id or drivers license? Do you have social media? Youre already in a data base.
@lobabobloblaw 25 днів тому
Well, I guess DLC content is going to get a whole lot spicier in Gaming Land.
@eriksilva631 25 днів тому
The guys that work to bring new compression/faster models are geniuses my hats off to them, any time I see a project like 1 mb fully working OS or something like 100x less data required, I appreciate it.
@conradjustinart 23 дні тому
I think I would rather see pixelated and compressed image of an actual face of a person rather than seeing their near-perfect model generated from a single photo from the past. Feels more genuine.
@r..k.. 25 днів тому
There is another important use case - the camera is often on the wrong position, even worse with a 2 display setup. This technique helps to look directly in the face of the others - that is one great benefit
@daveinpublic 25 днів тому
The first law of papers… Someone else will do the hard work in the next paper.
@khaitomretro 25 днів тому
Theranos
@ScenariosInAI 25 днів тому
A Day9 fan I see
@Antsaboy94 25 днів тому
Since the papers at 5:46 and 5:53 are nine and eight years old respectively, I wonder what more recent papers have to offer in video stylization. I remember that one paper that didn't rely on ANNs, but that required at least a single frame to be rendered in the target style. And if the character were to turn around, for example, the newly visible side would require its own sample.
@sky01james28 23 дні тому
Did you miss the video about Sora, a new video generating AI?
@vaendryl 25 днів тому
5:35 looks like a preview of the next generation of Vtubers.
@DuckieMcduck 25 днів тому
vtubers are a completely different niche. this is probably for automating youtubers in general
@vaendryl 25 днів тому
@@DuckieMcduck all you need is to change the output of the style transfer to "anime girl" and boom, vtuber. and without the expensive rigging required for live2D and with a lot more expressiveness.
@satibel 25 днів тому
@@vaendrylyeah, needing one drawing, instead of a ton of different drawings and time to stitch or a 3d model, cuts down the cost from like 500-10k down to 10-200 bucks*, that would make it a lot more accessible. *Basically it could turn a 1-3 pose png into a fully rigged model. I think we'll see that running on phones in like 5-10 years probably Though I'm not sure that exact paper will be used, advanced style transfer seems to fit a lot better
@Interpause 25 днів тому
Talkinghead already exists.
@RubenMorenoMontoliu 25 днів тому
I can see how this can be used for security validations where one has to record a video on themselves reading some text, or video calls impersonating someone from accounting asking you to transfer some money urgently, etc
@raymond_luxury_yacht 23 дні тому
Yup. Guess what face ri face meetings making a comeback. Time to reinvest in commercial office space lol
@firstnamelastname-or4sn 25 днів тому
imagine playing a game with your friends and you can see their real time rendered faces on their characters
@juschilltech 22 дні тому
Things get more grim every day
@baddrivers759 12 днів тому
Great progress. Does it look like a perspective transformation is in order? It looks like the background is moving at the same rate as the foreground as it rotates. I think leveraging the depth of field information to apply perspective to the generated image would fix the "odd" rotation feel that each of the videos have. Thoughts?
@AdriaanPretorius 24 дні тому
In the 90s I had this idea: why not give people a choice when they rent a movie. You go to a DVD store (those were the top of the line still, no streaming), pick a movie from a screen, make a few choices like the protagonist, villain and which actor should play them, and the shop would generate the movie from a master that was shot like any real movie, but with the new actors rendered in. So, if you like Top Gun, but not Tom Cruise, you could rent yours with Brad Pitt in instead. You would give the DVD back and they would erase it and re-use it for the next guy. It was a ridiculous idea.... right...?? It's a great time to be alive!
@SJ-cy3hp 2 дні тому
Had the same idea except it placed me and my friends and family into the roles with a simple voice capture and three perspective selfie’s. Then we watch the film together. Arguments like, I’m Gandalf this time! 🧙♂️
@aleks.l394 25 днів тому
for some reason it seems much more mind blowing, than even hyped claude 3 or stable diffusion 3. Truly, what a time....
@wwk279 25 днів тому
This is awesome tech but im afraid of personal security issue
@garystinten9339 23 дні тому
Eventually it'll be texture cloning.. as in as much as an image is an image.. detail in texture will be the next big thing.. meaning it will simulate texture also to get around biometrics.. and I imagine eventually sweat simulation based on texture of surface.. so frost on a glass melting in the sun will become a thing..
@zillboy 22 дні тому
Does this only work for person faces, or can we use it for other things as well like a landscape or car image. Thanks
@A.R.McDuck 24 дні тому
This is mainly good for the person who received the output, as they move around your output would shift correctly in their viewpoints.
@HorrorRod 25 днів тому
90's movie Ghost in the Machine vr arcade used the kids faces in game just like this, all animated of course for the movie but finally its here, star citizen animates ur face with web cam but does not do any of the face capture stuff like this.
@DracoTheNinja 24 дні тому
Holy crap, what a time to be alive! I cant wait until someone makes a game that takes just a view photos of ur face and an ai that uses lip sync-to-audio this is amazing
@goetterfunke1987 25 днів тому
Have a great time travelling to SF! Thank you for sharing the knowledge, i love your videos!
@mihirvd01 25 днів тому
4:46 Emraan Hashmi ffs
@_yuri 25 днів тому
hold on to your papers 😢
@AshwaniKumarSrivastava 25 днів тому
😄
@Brahvim 24 дні тому
I thought it was Himesh Reshamiya 😅 Yes, I'm an introvert. I'm sorry.
@ericjorgensen6425 25 днів тому
I would take this as- is for video conferencing where it appears that we are looking directly at each other.
@kevinbrown7308 25 днів тому
Is it possible to download this and try it? I can’t seem to find the code but I know Nvidia sometimes lets you download software if you have their graphics cards. I currently have 2 4090s
@davehedgehog9795 24 дні тому
what's the software called and where acn i download it?
@janrajmont9991 25 днів тому
Is it possible to use this tool? Can I access it as a user?
@goi4246 21 день тому
AI taking over the world: " What a time to be alive! "
@jordanfry2899 7 днів тому
I just want to appreciate the fact we've gotten to the point where we can generate the data of a 3D model of a human being simply with a photograph of them
@zerocoll20 25 днів тому
This method looks like they set the basis of reconstruction in the tip of the nose and generate the rest of then.
@sky01james28 23 дні тому
No, that’s just the pivot point they used for testing. Other snippets showed unique positioning instructions
@ahegpbtrftcotu 23 дні тому
It's impressive enough on _one_ reference photo - giving it even like 4 from different angles, or (imagine, heaven forbid) a one-second clip of someone looking around, would allow for SO much more stability than it even has now, to perfect some of that uncanny depth oddness. This is going to be INCREDIBLE for minimizing data usage... as if we didn't just find a way to get a _million_ times faster transfer on fiber optic, heh.
@CreamusG 25 днів тому
It's interesting to me how this AI gives everything a significant overbite. I wonder why
@Littlefighter1911 24 дні тому
Sending keyframes like every 300ms, lerping them (or using some other interpolation), then only apply the extra data on it, yeah might actually be less data sent over a network (although I suppose the overall computational power would still be higher)
@imjody 25 днів тому
Absolutely insane! The progress is never-ending, and with haste, and I love it. 😊
@alemangioni 25 днів тому
Amazing what this technology will allow. Great paper Dr!
@jayairstuff 25 днів тому
Rainbow 6: Vegas had to ability to take a photo and add your face into the game in 2006. Not to mention a huge appeal for games is to NOT play as yourself and embody someone or something completely different. Its escapism. This all seems very gimmicky to me personally.
@Vynzent 25 днів тому
A ton of casual RPG players gravitate towards simply making the character look like themselves. I wouldn't say it's a gimmick.
@H-famTV 13 днів тому
Lol, The old lady model at (2:35) ...is actually my late grandma wow, imagine my shock seeing you turn her into a 3D Avatar wow. I wanna do exactly that for my gaming project? 😇
@melonga7435 25 днів тому
Y'know what this will be cool for? Anime. The fact that it can turn a 2D stylized image into a 3D model means you may be able to get 99.99% accurate 3D anime adaptations in the future.
@Sergeeeek 24 дні тому
It would be like trying rotoscope animation in anime. It wouldn't look very good.
@MaxHarden 23 дні тому
Was ready for the heads to start spinning around
@deondrex18k 22 дні тому
Imagine how crazy this will look like in the next decade 😮
@thelazy0ne 24 дні тому
Finally I will be able to have my favorite adult star in my favorite adult game 😁 Much blessings to de paper writers! Please continue 😁👍
@jasonhildebrand1574 13 днів тому
I had wanted to simulate reality with a complex physics and chemistry engine for my new voxel VR game idea, but now it just makes more sense to not simulate anything and just generate everything in realtime with AI, amirite ?! lol
@thalleslucas2408 24 дні тому
this is new level in this techno-magic system
@marianoescalera3086 17 днів тому
how could I run it in my computrer?
@dhudhwani 25 днів тому
4:58 bruh 😂
@TreeYogaSchool 25 днів тому
This technology continues to get more and more impressive. Thanks for the great video.
@ethzero 25 днів тому
Low bandwidth but very GPU intensive on the client side. Hopefully this can be improved as it wouldn't have much of a use case for mobile.
@TechnicolorOutlier 25 днів тому
gforce now
@exosproudmamabear558 24 дні тому
Machine learning technics gets optimised fairly well. They get smaller and faster with every paper so we may make it work in a few papers after fpr mobile usage.
@praveshsingh8630 23 дні тому
Nvidia AI + Apple vision Pro = Magic 🎉
@Entropy67 25 днів тому
I was waiting for this one 😁
@raymond_luxury_yacht 23 дні тому
So how do I demonstrate it's me talking and not a hijacked version?
@adispadi4820 25 днів тому
What the Heck. Just today I sat on the wall you See in the picture at 5:45. I am pretty sure the City is Tübingen, south of Germany. What a coincidence to See it in a twominutepapers video!
@handlenotset 22 дні тому
how do i try this out, anyone know?
@andersonantunes7621 25 днів тому
Im a little bit concerned, all Brazilian online banks uses face based validation for security and it uses pictures instead depth sensors. I hope they have another idea on how to fix this
@mxmstrj 22 дні тому
Unexpectedly seeing Coleman Hughes in a paper like this
@dvvinever 23 дні тому
Midi for music Vidi for video? Just send the instructions and Ai/hardware will recreate the scenes and eye candy. The day when a kid can create a movie quality AAA video game in one minute at home just by typing in a prompt is coming faster than we realize. Just imagine what will be possible in 10 years time. Will UKposts take advantage of this in the future and reduce videos down to instructions that have a user's video card and processor recreate a video? Could future stored video be reduced in size to unbelievably small sizes saving billions in storage costs and also reduce the strain on networks?
@percywhitehead9228 25 днів тому
it's impressive it can do that on commodity hardware but it's like nothing compared to EMO by alibaba, which not only looks more realistic but it also changes the facial expressions and lip syncs to anything you give it even video. This is disappointing compared to that
@smellthel 25 днів тому
Ultra realistic Saul Goodman 3D
@AurrenTV 23 дні тому
2:23 lol no they aren't, it's just a texture plastered on front, they dont change as it's rotated.
@ShadesOfMisery 25 днів тому
lol the Day[9[ cameo xD
@SP-ny1fk 23 дні тому
Imagine if your idea of amazing is to be able to watch UKposts all day
@27Jaak 24 дні тому
In before people will abuse this to successfully blame crimes on less fortunate people. If these research papers keep evolving at this rate, this is not very far from reality. We need to reconsider many things in our society or else our current society will collapse. What a time to be alive! Your channel is more designed for infotainment, I get that, and I thank you for broadening my world view!
@TheStrandedAlliance 17 днів тому
I wonder how long it'll take until these techniques are actually publically available...
@istiblasko1 22 дні тому
See you in San Francisco! Can't wait!!
@KootFloris 25 днів тому
Karoly, I think it's also time to make a few videos where these things can go wrong of criminally be abused. Like face recognition to open other peoples phones. And worse. Too many tech advocates applaud every new possibility, yet as with cars, planes, batteries, internet, they all have dangers and new problems too.
@justinwescott8125 25 днів тому
Like you said, every single new tech can be abused, and we all know that. Not much point in ruminating over it. Plus pessimism isn't the vibe of this channel. There are plenty of other channels you can go watch for doom and gloom.
@KootFloris 24 дні тому
@@justinwescott8125 The positive vibe is a reasonable point. No doom and gloom thank you. A bit of wisdom around tech that may help never to go there, might be fun though?
@_sky_3123 25 днів тому
Would love to have this implemented in character creation for AAA games. If I can create a character, why not have one looking like me.
@RaySquirrel 25 днів тому
4:39 Is that Coleman Hughes?
@hbasm3271 23 дні тому
What a time to be alive, when the world is about to buckle under the pressure of this new tech. Watch it all come down.
@southcoastinventors6583 25 днів тому
Only need a top of the line Graphics card to get it to work, he must have a tinybox as his desktop.
@honoiki3778 25 днів тому
What a time to be alive!!1 1(idk what to say, but I want to be one of the first people to say something)
@noinghenah2764 23 дні тому
=) how much energy does it take?
@sirkingjamz101 25 днів тому
PH is gonna need a new category soon: AI
@epicthief 25 днів тому
1 step closer to Harry Potter magic photos
@earFront 25 днів тому
Thank you for all the updates on what va wonderful world we are creating.
@AllisterVinris 25 днів тому
So automatic Vtuber model rigging ? Awesome!
@realWorsin 24 дні тому
I didn't even know I needed to put myself in a video game. Learning all sorts of things today.
@joelcoll4034 25 днів тому
It looks awesome!
@t1460bb 25 днів тому
Do you know what this article means, my friends? In the near future, all video games will no longer require hours and hours of work to model characters and objects. We will have extremely realistic games with much less work. In a basketball or football game, for example, we will have virtual characters exactly like the real players! This is incredible! WHAT A TIME TO BE ALIVE!
@sapien82 25 днів тому
cant wait to just make 3d models from my 2d character sketches so cool
@kras_mazov 22 дні тому
No way i'm talking to my beloved ones using that.
@joakimawesome 24 дні тому
Isn't that guy Marques Brownlee (0:04), the UKpostsr who does product reviews?
@TechnicolorOutlier 25 днів тому
now we need some papers on the sociopolitical implications of this technology to accompany this one
@rickymort135 25 днів тому
7:00 Wait I thought weights and biases was the best....
@L_tlu 25 днів тому
same
@issacdhan 25 днів тому
I Invested 2 years to learn facial modeling (2002-2003). Then photogrammetry became popular . And now it's done in 16 ms. OMG. Worried about 3d Artists.
@michalpetrilak3976 20 днів тому
OVER GAME already applies to 3D artists. Everything will be pushed through A.I. graphics products because it's faster and cheaper.
@danko5866 20 днів тому
4:58 Immran Hashmi
@aliancemd 15 днів тому
“Technically” does what is advertised but the output is looking so weird that it’s unlikely to be used, unless you are fine with creeping people out
@piotrmorag2597 25 днів тому
I wonder what would happen if we were to show it a mirror 😅Like the guy putting on headphones in real time, but with a mirror instead.
@Vikcreed 25 днів тому
this is insane. A person IN REAL LIFE could hand, or pass an object to a person VIRUALLY by rotating in briefly in front of the camera! all object can automatically become virtual. In fact this will be able to map the planet earth virtually. This could go full brain in the jar technology fast!
@yakirfrankoveig8094 25 днів тому
It seems it has an issue understanding that the cats body also needs to be rotated
@Yantryman 23 дні тому
Can it see my face from seeing my ear? And big up to a Doctor with Hungarian roots!
@foxriver9156 21 день тому
Hello 4090, goodbye boring Zoom meetings. Though I will need to hire a PA from overseas to be in my meetings.
@badizzl 24 дні тому
As awesome a these advancements are, I'm still not comfortable with the rate of progress in AI, because all I can focus on is the stuff bad actors can do with these technologies, while most people (even in developed countries) barely know what ChatGPT is
@ApEsXiT 25 днів тому
What a time to be alive!!!!
@UgalabaguUgabagu 24 дні тому
4:49 Emran Hashmi!!!
Filmegator
Переглядів 1,8 млн
Жекич Дубровский
Переглядів 3 млн
The Wall Street Journal
Переглядів 1,4 млн
Сэм Альтман рассказал о GPT-5 и будущем OpenAI | Роботы Tesla вкалывают на заводе | Робопчелы и Марс
PRO РОБОТОВ
Переглядів 54 тис.
PRO Hi-Tech
Переглядів 88 тис.
i-shoppers обзоры
Переглядів 63 тис.
Rocket Tech School
Переглядів 52 тис.
ProTech
Переглядів 40 тис.