AI Characters: Animation, Consistency, & Video!

🌏 Get an Exclusive NordVPN deal here ➼ nordvpn.com/mattvidproai
It’s risk-free with Nord’s 30-day money-back guarantee! ✌ Don’t forget to use my discount code: mattvidproai
Thank you NordVPN for sponsoring this video
▼ Link(s) From Today’s Video:
Open AI CTO on Sora: / 1768347097324257423
Claude 3 Haiku: / 1768018310615151002
Creepy Open AI Robot: / 1767913955295744449
Dogan on MJ Consistent Character: / 1767331560632832441
Emmanuel on MJ Consistent Character: / 1767343807476170878
Mckay Wrigley 27 min Devin Demo: / 1767985840448516343
Cartwheel AI Animation: / 1768312252141273157
Gemini 1.5 is a bust: / 1766862660677648695
Nvidia RTX4080 super contest: • I'm Giving Away an RTX...
► MattVidPro Discord: / discord
► Follow Me on Twitter: / mattvidpro
-------------------------------------------------
▼ Extra Links of Interest:
✩ AI LINKS MASTER LIST: www.futurepedia.io/
✩ General AI Playlist: • General MattVidPro AI ...
✩ AI I use to edit videos: www.descript.com/?lmref=nA4fDg
✩ Instagram: mattvidpro
✩ Tiktok: tiktok.com/@mattvidpro
✩ Second Channel: / @matt_pie
-------------------------------------------------
Thanks for watching Matt Video Productions! I make all sorts of videos here on KZbin! Technology, Tutorials, and Reviews! Enjoy Your stay here, and subscribe!
All Suggestions, Thoughts And Comments Are Greatly Appreciated… Because I Actually Read Them.
-------------------------------------------------
► Business Contact: [e-mail]

[@MattVidPro]

🌏 Get an Exclusive NordVPN deal here ➼ nordvpn.com/mattvidproai It’s risk-free with Nord’s 30-day money-back guarantee! ✌ Don’t forget to use my discount code: mattvidproai Thank you NordVPN for sponsoring this video

[@LouisGedo]

👋

[@LouisGedo]

Technology is never intrinsically good or bad; it's always HOW technology is used that we can assess the goodness or the badness of that technology's use.

[@NicVandEmZ]

Ive been talking to pia ai more than ChatGPT

[@skiroviepcvr7792]

1:06 Hey! What about this totally free VPN: Proton VPN and Orbot NordVPN is not totally free to use.

[@mycelia_ow]

The link to the NVidia event is not in the description btw

[@timduck8506]

Based in Panama: While NordVPN operates outside of the jurisdiction of major surveillance agencies like the NSA, its headquarters in Panama could potentially make it subject to local laws regarding data retention.

[@googleSux]

Panama was once invaded by the US. Nothing else to add to that.

[@elirothblatt5602]

Your videos are great and I love the topic. Thanks!

[@Mattorite]

They are really going to need to make the models more efficient. It has to be extremely expensive to send API calls all day.

[@BytedDev]

Love you Matt ❤

[@ran4erep]

marry him

[@hipjoeroflmto4764]

Have kids with him

[@vi6ddarkking]

The funny thing about competent AI agents. That can program really well. Is that they won't last long. Not because they aren't good. But the moment they can. The open source AI agent developers, will use it to catch right back up.

[@jaredf6205]

Midjourney's style reference mode was also just updated today. Works really well now.

[@yoagcur]

Not for me. Even at --sw 1000. Very disappointing and a backward step from my attempts

[@yoagcur]

Actually a bit hit and miss . Works with some styles better than others. I think the color is dominant

[@kevinkingmaker7395]

"Not sure" if Sora used KZbin videos! She is worried about liability.

[@dbt4869]

Are you blind or just stupid?

[@jeffsteyn7174]

They have a deal with shutter stock and she wasn't even certain about that. It's very unlikely for a cto to know the details of what data was used on a day to day bases.

[@juanjesusligero391]

​@@jeffsteyn7174I'm pretty sure she knows very well whether KZbin or Shutterstock videos are being used. She just happens to also know that it's not a good idea to give that information to reporters XD She was probably worried about liability and also about letting the competition know anything related to their training data.

[@IceMetalPunk]

Mira answered the question about training data as best she could, since there is no right answer. People who incorrectly believe that training an AI means, basically, "storing the training examples in a database for the model to copy and steal later", will never be satisfied. She could have said "we only trained it on data we produced here, in-house at OpenAI" (never mind that wouldn't be nearly enough to train Sora anyway), and people would still be outraged that they "exploited their own employees' creative labor and stole from their own workers!" The people who believe AI is a copypasta machine don't want the truth, they only want to hear one answer: "We haven't trained the model on anything because we now agree with you that all training is theft, so we've decided to stop making AI models and shut down the organization entirely."

[@Halo6166]

Wow , such an amazing response . This is exactly how I felt , and it truly enrages me the stupid politics revolving training data , my god just let these people be . As if it learning from public and available information is the worse of the worse kind of sin. Humans are just coping

[@SebaBuenoHaceMusiquitaJijiji]

dont ask monkeys to understand the future, we cant, unga bungaaaaaaaaa

[@markonfilms]

Yeah it's so silly and it's a lost cause. I mean, don't people realize the wealthiest already have everything? They don't need AI. We need AI! If they were just trying to replace everyone for themselves they'd have already done it lol. Most goods and services are consumed by most people. I'm a lefty and have mixed feelings about corporate profit etc. This may free us from labor largely, and hopefully the poorest people in the world will be lifted up too. More time for enrichment, health, and education/learning and growing as people.

[@virgilxavier1]

They won't be happy with that either. They will then want vast reparations for wrongs already done that continues for centuries and that generations not yet born are obliged to pay.

[@jeffsteyn7174]

​@@Halo6166next we going to hear we need to pay copyright for reading stuff online

[@Yipper64]

6:20 when I ask claude questions without giving it documents to base my questions off, I have noticed it has a tendency to hallucinate more than chatGPT. Which is interesting, I wonder if there's like a push/pull of diminishing returns, like if you have an LLM that has great recall for text its given, it will hallucinate from its training data, but if its so-so on that, it can be a lot less likely to give you wrong information from its training data.

[@retrofuturism]

Information Kintsugi Inspired by the Japanese art of Kintsugi, where broken pottery is mended with gold to highlight its history, information kintsugi involves AI enhancing and reinterpreting human-generated data. AI fills in gaps and adds its unique creative touch, transforming the original into something new and valuable. AI becomes a collaborator, fusing human ideas with machine aesthetics.

[@hurktang]

@mattvidpro , They already said Sora was able with very little fiddling to play videogames such as minecraft. They might have trained the robots in the same way. That is: making the robot act in a way it look like "a robot handing an apple to a man" using sora and probably gpt-4.5 as the LLM

[@agnesslovehealz]

If I entered I found my conf can i resubmit ?

[@MrRandomPlays_1987]

8:00 - how the world would look like in 30 or 40 years ? wow, let alone even just 20 years into the future and we would have a very very different and much much more developed world (if all goes right and current quick pace of AI developement keeps iterating and progressing forward), according to ray Kurzweil, the (true) singularity should happen in 20 years (in about 2045+) so even by then the power of AI would be so immense and unstopable to the point of having a singularity event or something far crazier than whatever we saw up to this point and will see in the future, hopefully it would be a positive change for making our world a better one on every aspect, by 2045 an advanced ASI AI system should exist by then which should be able to solve and cure all or most diseases and be able to solve age reversal tech (if by then previous AI systems like AGI and humans wouldn't be able to yet) the ASI would create new sciences and would invent new technologies and would solve the hardest of problems in science and much more, and so it's worth waiting for.

[@countofst.germain6417]

Good vid Matt, it was disappointing that openai didnt even acknowledge the one year anniversary since the release, at least make a tweet ffs. Eitherway I'll be moving over to Claude.

[@NickDrinksWater]

Sora better be staying hydrated

[@Deckardb25364]

Have no useful comment, just putting one for engagement. Yaaaaaa

[@user-uq1sn5ob3k]

Hey Matt, can you make a video illustrating the best AI tools currently available for each tasks (image gen, animation, simple coding)? Asking cuz there are sooooo many >;( . Luv u

[@chin-qo2tj]

the also need to give us consisent characters like RIGHT NOW, with dalle3 as its clearly capable of doing do

[@MilesBellas]

WSJ published a negative article on Emad.

[@hypercoder-gaming]

W matt

[@InspiringWorship]

I found out it’s nearly impossible for AI image generators to do images of people with teeth gaps. I hope they fix it!

[@seikojin]

So Open AI contracts with Data source providers. Those data source providers do scraping and have less stringent controls on obeying laws for fair use.

[@TiagoTiagoT]

Woah, she seemed really uncomfortable about answering that question about the source of training data for Sora... Should've practiced with their lawyers more before going on record about these kind of things.... Personally I wouldn't jump to the conclusion they broke any laws; but she was definitely not prepared to be put on the spot and ended up making themselves appear worse than they probably are...

[@boozonggy]

she kinda looked guilty

[@TiagoTiagoT]

@@boozonggythat can happen even when innocent people get nervous, specially during interrogation-like situations. Right not there's not enough information to know which way the truth really goes here.

[@boozonggy]

@@TiagoTiagoT ya

[@boozonggy]

ya@@pattyprolapse

[@TiagoTiagoT]

@@pattyprolapseIf it's "publicly available", then by definition it isn't stealing; it is available to the public, they are part of the public, therefore it is available to them.; very simple logic.

[@dnero6911]

NVIDIA Chat with RTX is SUPER SUPER SUPER Cool ...

[@JohnSmith762A11B]

What? It's so lame compared to something like Faraday.

[@thesource0024]

video quality 360p, wtf?

[@MeNoOther]

Consistent characters and brand names mean no more hiring photographers and models for ads

[@hombacom]

More of unrealistic stereotypical models you mean

[@scrutch666]

Lol the character in this mj consistency test looks completely different besides the hair color

[@kuromiLayfe]

if gpt 5 releases , i hope gpt 4 will be the free default instead of 3.5 turbo

[@BackTiVi]

That would be cool, but i doubt it considering gpt4 is much more costly at inference than gpt3.5

[@archdemonplay6904]

you can already acces gpt4 for free in microsoft copylot

[@apache937]

@@archdemonplay6904its trash there

[@MrNewBanks]

I can't believe it's been a year.

[@carlkim2577]

She can't say other than 'publicly available' because they are getting used. Don't blame her for that. It's for the courts to work out first.

[@chris_jorge]

used or sued? it must be a minefield to navigate research, prototyping, and commercializing these products.

[@lucamatteobarbieri2493]

So sad thad ClosedAI it is closed sourcing even the robotics new stuff now.

[@AiandMusic]

I think it would be genious to Open AI - and to any similar company - to NOT know themselves what data was used to train the model. Not sure if it could even be possible, but just a thought. They could literally pass lie detectors etc. and just say in court: "We or no one on this planet, knows how the data was trained.There are no logs. Ask Sora." Or if that's actually how it Is right now.... "Well hello AGI" :D

[@sneedtube]

Those robot movements feel hard-coded so I don't see anything new, they just put old OpenAI's stuff in it. Pretty sure a similar demo could be made by a script kiddie with a Raspberry Pi

[@RealShinpin]

bro you need to grow a beard, you would look good with one .

[@puddingthes_j.wde5t.r0y3r4]

NICE NEWS

[@Pauluz_The_Web_Gnome]

NORDVPN!

[@SuperVillainRetroFace]

Keep the videos coming the super villain needs Deceptikon technologies to take over the world good work matt

[@rodentrider2]

hi

[@christiandarkin]

of course they used youtube for sora. of course they did. there just isn't enough video anywhere else to train a model.

[@luckyme5122]

Awesome… Roll out of bed make a KZbin video without combing your hair

[@BodaneNippleRaider]

And

[@sveinnrafnsson851]

MJ consistency parameter doesn't work for clothing so its useless.

[@tedstriker2000]

it bugs me about where 'data' comes from - this then means Photo montages are all stealing. If I'm inspired to create something similar - am I stealing. If someone draws a bird - Can nobody never draw a similar bird in a different style or pose - its been done. Its dumb. Ive yet to see an AI image that stole someone's image from Deviant Art and put it out as new.

[@sebastianmuller1624]

9:20 Haha, what, the X (broken Twitter) Video is not playing? ... Open Source Genius Musk (don't delete Musk bashing comments)

[@jashall30]

Why dose this video look so old 😂

[@mwdrum]

KZbin is public for public viewing and consumption, not stealing.

[@southcoastinventors6583]

Stealing free video ?

[@mwdrum]

She’s lying…..total bs.

[@lp712]

Even if she was … who fucking cares….. quit crying. Training is NOT stealing. Especially if it’s publically available free content….

[@Airbender131090]

SORA is a scam. Not gonna be released till 2025 and will not have the same quality they showed off at 1st presentation. Sora team says release is not anytime soon.

[@GaryJr530]

This comment was generated by a human 🙃

[@nickgirdwood3082]

Wall Street Journal. Well there you go. Terrible company.

[@gagrevolver]

weak video matt. didn't even test mj's new feature just regurgitated what a competitor said. you're slippin dude.

[@MattVidPro]

This is a snippet style news recap not a review of the midjourney feature. If I decide to do a video on midjourneys new feature it will be a lot more in depth with a comparison to other consistent character strats..

[@gagrevolver]

@@MattVidPro Firstly, sick to see a response, thx. I only say this shit to prod for better content from you so no hard feelilngs. im a dick with a purpose. Secondly, idk man. itd just be nice to see some cool stuff from you instead of you regurgitating things 12 other people already beat you to the punch on.

[@fizzypizzel6477]

W matt