Even More MASSIVE Video AI Upgrades & New Models!!! it just does not stop!

[@IanHollis]

At this rate, OpenAI's Sora's going to be redundant by the time it's released.

[@johntnguyen1976]

100% already is LOL

[@VisualVortexArtStudio]

Hopefully they make it well priced or included with the normal gpt subscription which would be great

[@vainezaiven6677]

Sora will never be released to the public.

[@vargskelethor]

wouldnt be the first time that happened

[@brianjanssens8020]

Sora's releasing this week or next week probably.

[@Atheist-Libertarian]

6:20 😂 Unparalleled level of understanding of physics

[@jamesvictor2182]

I'm so glad you pointed out the hair. I hadn't noticed at all strangely

[@DJ-Umair-Ali]

6:19 very impressive physics

[@jrikkerink]

Came here to write the same thing, lol🤣

[@AmandaFessler]

Gotta secure the ballistics.

[@So_-.]

This is getting scary

[@Artorias920]

BRUH, why are we on the same page LMAO

[@LucasAndradeX]

Hmm yes physics

[@Vine_Zx]

Jiggle physics are insane.

[@MattVidPro]

Open Source FOR THE WIN!

[@Zuluknob]

*80GB of vram required.

[@TPCDAZ]

and totally pointless as only studios can afford the cards needed for it to run. It's like giving away free money but making the box it's in so heavy only Thor can lift it. Whats the point

[@armondtanz]

​@@TPCDAZor Elon saying win a free cyber truck then saying congrats ,, you won , u can go collect it, its 2000 miles into space.

[@AgustinCaniglia1992]

@@Zuluknob Lots of open source models start at high vram requierments and then the community does optimizations and you can run with potato.

[@Zuluknob]

@AgustinCaniglia1992 not 80gb ones

[@CodexPermutatio]

Open source video generation. This is the way!

[@vomm]

Yeah we need this so bad as the commercial video generators are just insanely expansive. I hope so much we see video gen for consumer hardware soon.

[@issabarbier2139]

Matt: Speaking about physics on the body and hair movement. Me: Looking at the woman reflection on the glass window in the corner.

[@CodyCha]

Exactly. That's the first thing I noticed

[@therandommusicguy4773]

You are arguably the best AI news source, just because you only give things as much hype as they need. Thanks for being real man

[@jeffwads]

Dude, in 10 years, all of us are going to be strapped in a Matrix-Lazyboy exploring fantasy worlds.

[@shortstoryuniverseaudio]

No joke.

[@FilmSpook]

@jeffwads BS. Remember 2020?? Well, that was Disneyland compared to what is coming.

[@FilmSpook]

@@shortstoryuniverseaudio BS. You seem to have forgotten 2020. That was "good times" compared to what is fast coming, son... Accept reality.

[@Raulikien]

99% my prediction

[@brexitgreens]

I already am. You wouldn't believe the things I've seen.

[@역행자-n2f]

tbh only reason I click on your videos for AI news instead of other channel is because you don't do the SHOCKED EYES WIDE OPEN, MOUTH GAWKING 😲😲😲😲😲😲😲😲thumbnail face

[@Adam-dz5tj]

Exact same thought process 😊

[@tuckerbugeater]

Half of his videos are a waste of time

[@caligoclarus]

@@tuckerbugeater Now we are left wondering who you are talking about.

[@therandommusicguy4773]

EXACTLY. This guy never does clickbait even in a minor form. Other's like Wes Roth seem to hype up things which aren't that good in their video titles to gain clicks and it makes you feel like you never know if a video is gonna worth your time.

[@vomm]

@@tuckerbugeater that's only half the wasting of time as reading your comment

[@High-Tech-Geek]

Please specify at the BEGINNING of each segment whether you're presenting just a paper, or an actually available tool and whether there's a demo page or a subscription or if it's open source or requires ControlNet or ComfyUI, etc. I have no interest in vaporware or papers that tout upcoming features we won't see for a year.

[@High-Tech-Geek]

Tencent states that Hunyuan requires 45GB of VRAM for lo-res and 60GB for medium res but they recommend 80GB. The Nvidia A100 80GB generally costs $15,000 - $20,000 just for the graphics card, so...

[@brexitgreens]

Why do people suddenly expect to be able to run everything at home? There will always be something bigger than you can handle. It used to be an obvious truth - not anymore.

[@High-Tech-Geek]

@brexitgreens the entire point of open source is so people have control and can run the stuff themselves. This was released as open source.

[@brexitgreens]

@@High-Tech-Geek Cat mentality. Open source is not just for individuals. Themselves ≠ alone.

[@jamesvictor2182]

Matthew Berman got sent a 96gb dell to play with. Why haven't you got one? Not fair!

[@JonnyFootman-k5t]

RAM or VRAM? those are two different things

[@brexitgreens]

Who will foot his electricity bill though?

[@jamesvictor2182]

@@JonnyFootman-k5t Thank you for pointing this out. I agree it may not be clear to some I was referring to an AI capable workstation in the context of being able to run larger AI models. The machine I refer to is a dual RTX a5000 AI workstation with 2x48gb of VRAM

[@jamesvictor2182]

@@brexitgreens A typical user of dual GPU systems will not see much impact on the electricity costs, since it's likely usage would be sporadic. Also the power draw of an a5000 is a modest 230w max, so anyone with a modern GPU system wouldn't see much increase in bills

[@brexitgreens]

@@jamesvictor2182 Thanks.

[@maximumblackness]

7:48 dude looks like Obi Wan LOL

[@BlackMita]

Anime is so easy to ‘fit’ I get reverse impressed when a model’s demos feature it.

[@Justin_Arut]

Sora who?

[@NLPprompter]

cmon.... sora wasn't released because they pressured not to especially those Hollywood folks, but see what happens when they try to surpress AI models release they'll missed out the hype, to me over here i saw Kling win because of this stupid politic. especially when they say worry about it could be used for negatives things in the presidential election, guess what Elon did that and everybody does too without sora. not releasing sora was really baaad decision, if you don't have compute in this era... you open source it, or release the paper of it at least you'll get the recognition rather than keeping it and letting other have it without get anything, pfft they even got slammed by some artist saying artist is not beta tester or something.

[@SinCityRadio2077]

Last vid we got the perfect black leather pants and on this one accurate physics at 06:19, the minimax new model is also perfect for my use case

[@AngeloWakstein-b7e]

Awesome as always!

[@bobbykingAiworld]

Hello Matt! Great video keep up the good work! 😆

[@isajoha9962]

Many cool leaps in video generation.

[@KBStories60]

Finally!!!! Thanks Matt!!!!

[@fragglespank]

"Twelve vee oh one live" lol, it's an I mate. As in "image 2 video"

[@ElnoraPerkinsns]

Your videos are always so informative and informative! Thank you for this! 🔥😛

[@matt.stevick]

more treadmill videos…. asking for a friend.

[@nexyboye5111]

the bottom woman is more like a creepy doll

[@brexitgreens]

I prefer the top too. Natural. Sadly blurry.

[@TLabsLLC-AI-Development]

This will all automatically apply to the Temporal Prompt Engine too. I'm looking forward to developing tonight.

[@VFXShawn]

Need a tutorial on installing Hunyuan Video to work with ComfyUI and image to video workflow.

[@matt.stevick]

god some of those videos are so gorgeously perfect 👍🏼 📈🇺🇸

[@peacetoall1858]

For now one maybe able to run the Tencent's Hanyuan video model on RunPod's A100 GPUs with 80 GB VRAM plan

[@dwainmorris7854]

Still these jewels need to have more control as an artist prompting is not going to be enough As an artist I don't have to design characters from the scratch and from every angle

[@J0r1ckV]

6:30 Look at bottom left, the reflection of the woman in the window

[@harnageaa]

I want to specify something . You're wrong about in-between frames for motions, Saw the nitpicking that they skip frames but most anime are 24 frames aniway. And many times most movements contain frame skipping or things looking weird. Artists save time by drawing jibberish in movements to save time and they only care to look good when it's still. The only time where you get fluid constant frame is in movies or in high budget episodes or only certain scenes. So in my opinion it looks beyond fine. It's insane. If not for the morphing and the "hands" (always the hands) I wouldn;t be able to tell it's A.I at all. EDIT: I watch about +10 new anime every 3 months, so about 40 new anime per year on average . And after 13 years of anime I can say that animation i've seen there is better than 80% I usually watch, which is mind blowing. Can't wait for even mid shows now to look like jujutsu kaisen or demon slayer in the future :O.

[@Role-Play_Gamer]

What about Dragon Ball Super? Nothing can beat that top tier animation.

[@viperjay1]

Did I spot a quest 3 in the background ?

[@guitar_investment_money]

massive co2 on the atmosphere

[@MrDonCoyote]

@5:39 unless you're on a Mac with a metal chip, where ram and vram is essentially the same and allocated according to use case. My Macbook Pro has 128 gb RAM, which means I can run it no problem.

[@mylittleheartscar]

Which mac?

[@MrDonCoyote]

​@@mylittleheartscar I have a macbook pro m3 max.

[@brexitgreens]

I find it hard to believe. Why would anyone be buying Nvidias A100 if the same could be accomplished by buying a Mac?

[@MrDonCoyote]

​@@brexitgreens The use case between the two cannot really be compared. The NVIDIA A100 is a specialized GPU designed for demanding computational tasks, particularly in the realm of deep learning and high-performance computing. AI generation doesn't require this unless you plan on training your own models or something. But for AI image/video generation, especially with those high-res models, you need a ton of RAM. That's where the MacBook Pro with 128GB shines. Plus, Apple Silicon chips are basically made for macOS, so you get that extra performance boost.

[@brexitgreens]

@@MrDonCoyote Awesome if true. I will ask Perplexity to verify it but now I need a nap.

[@tedstriker2000]

Id like to be able to see were the images came from and how much - - - If Im just getting a montage of Stock images, Id feel a lot less creative

[@TLabsLLC-AI-Development]

Matt. I will work with you to help provide access to these higher end gpu instances through my Microsoft sponsorship. Please feel free to reach out. I have most of it setup already.

[@TheWizardsTales]

Are you willingly lending your voice to those NOTD commercials or are they deepfaking you?

[@channelveritas7474]

2nd!

[@CaritasGothKaraoke]

“dynamacy”?

[@hipjoeroflmto4764]

Shiba hub shiba hub moment

[@kd-zq5oq]

You need a bobble head

[@SmartGR2014]

Guys I'm looking for local image and or video generation without restrictions on my pc, which do you suggest? Heard of pinokio, comfyui, invokeai, nmkd, a1111...what u suggest?

[@BrianWaltonDeveloper]

I’d suggest Stability Matrix as a base as it’ll manage all of the uis and even let you download models. It even has its own inference page for quick image generations that piggybacks on ComfyUI. ComfyUI is essential so I’d say start with that since you’ll end up having it anyways but you can add new uis so easily with stability matrix that you can just try them all and see what you like most. Hope that helps!

[@CALIODD]

Matt what do you think about ANTI-AI people? I wanna know what do you think about it.

[@aliettienne2907]

5:17 The VRAM is helping the video generation to be of stellar quality, the intricate and ample details are vivid and perfect. Video effects are smooth and the dynamic is consistent and flawless. 11:07 The video animation generation is spectacular and the dynamic rendition blends perfectly in every transition. The animation realism is surprisingly convincing and the artistic quality is vivid and smooth.

[@zsombi123]

first

[@realjgerard]

Virtual RAM

[@odrammurks1497]

✌

[@Andy-df5fj]

Will AI girls finally know how to walk on two legs?

[@mircorichter1375]

China open sourcing all the top AIs lately is not altruistic. Its a strategic move in the economic battle against the US, where the people just happen to benefit from... So Yay! Economic battle

[@brexitgreens]

"Competition benefits the consumer." Really??? 😳

[@mirek190]

lol ...cope like you want

[@elon-69-musk]

i love china more and more

[@TPCDAZ]

Tencent is good but requires an 80gb $40,000 gpu to run so pointless for most people

[@brexitgreens]

So things are useful to you only if you can own them exclusively? 🤔

[@TPCDAZ]

@@brexitgreens Your comment has quite literally nothing to do with mine. Not even close. What I suggest is you either delete your comment or re read mine and this time try engaging your brain first.

[@mirek190]

nah a100 80GB you can buy for 15k already

[@brexitgreens]

@@mirek190 Wow! Cheaper than a car!

[@TPCDAZ]

@@mirek190 If you have £15k to drop on a GPU you're in the same realm as someone who would drop $40k anyway. £15k would also either be a very well used one that already has a million hours on it from fantasy bitcoin miners or a "is it real or not" one from a small Chinese seller on eBay sure. But we're talking actually legal and usable here.

[@0sn1s]

🤣🤣🤣we are not thinking about anything

[@NakedSageAstrology]

I really don't know why I bother with channels like this anymore. Just because you SAY its open source, doesn't mean it is! I wish we had people who actually know what the hell they are talking about.

[@XXAI-1]

This is a means to attract people to try it out!

[@NakedSageAstrology]

​@@XXAI-1 It's as bad as reaction content. No substance, just mindless drivel.

[@henrythegreatamerican8136]

Why are you wearing pajamas with chunks of cheese on it?

[@Awatanation]

By "Safety" do you mean censorship? 5:55 If so, you are out of touch, my man.

[@tombradford7035]

Hate these video gen talk talk instead of letting us run the damn thing lol

[@jasonireland7175]

aaaa

[@Xandercorp]

Dude, could you please start with how much it takes to run and whether it's available already or what...

[@TheBigb1973]

Whats with the 🍋? You like lemons?

[@vomm]

it all began with the lemon

[@shinrinyokumusic1808]

@@TheBigb1973 it's cringe.

[@limyeeshin]

Hailuo actually pronounced as "Hai Luo" which means seashells in Chinese

[@FOX_ON_THE_RUN]

Third XD

[@RaymondMathewsws]

How do you feel about the importance of attention to detail when creating content? 💚🍓

[@SjarMenace]

What you mean eveyone is focussed on im g so no

[@dwainmorris7854]

It's nice I see you got a man's haircut the bangs was not working for you , made it look like you were trying to hide A receding hairline

[@NighttimeJuneau]

What the hell are you even talking about, dude. He’s always looked like a perfectly attractive guy.

[@billknight7342]

All these advancements convinced me I need to buy a Mac Mini M4 Pro with 64GB of ram... since it is shared between the system and video memory it can run much this crazy open source stuff locally.

[@zzzzzzz8473]

recommend to read more on specific use cases , while the unified ram is shared and can load for example big llm models , the specifics of image generation or video generation especially in implementations like comfyui may be tuned for specific gpu hardware mostly nvidia cards ( for cuda and tensor RT optimizations ) , and nvida gpus already support paged memory that can offload vram to normal ram ( at cost of speed ) . i dont mean to talk badly of the m4 i think the unified memory is awesome , and wish other hardwares would similarly decouple or unify so we could just buy more vram to scale , ive just seen so much of the optimization and implementations in the image/video space be not for mac . its a tough decision tho too as the next gen gpus specs arent increasing their vram enough to these demanding big neural networks .

[@billknight7342]

@@zzzzzzz8473 Yep seen that nvidia bias, just trying to break the PC bias until they decide to catch up hardware-wise~

[@mirek190]

64 GB is not enough .... also you need in your mac the ram as fast as possible ... ultra version has 800 GB/s that version is only reasonable version for AI. So ultra with 192 GB ... still very good prove even if you compare for nvidia a100.

[@billknight7342]

@@mirek190 What specific use cases come to mind at over 64gb? So far everything I have wanted to play with only pushed the 40gb mark. I have even seen cases of errors in code for systems over 64gb because so few people use them nobody fixed the bugs.

[@mirek190]

@@billknight7342 Check how many GB/s has RAM of that 64 GB version ... if is something like 100 / 200 GB /s is not faster what any generic PC with current CPU and ddt5 6000 interface. (gou will be wayyy faster rtx 3090/4090 as has around 1000 GB/s ) ... so you need really fast ram 800 GB/s

[@djpuplex]

60 gb of vram 🫨

[@LoraModelo]

Hi Matt, love your videos