OpenAI’s Sora Is Here, But…
❤️ Check out Lambda here and sign up for their GPU Cloud:
Sora is available here (for select countries, not in EU currently):
Earlier video looping paper:
📝 My paper on simulations that look almost like reality is available for free here:
Or this is the orig. Nature Physics link with clickable citations:
🙏 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, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Sundvall, Taras Bobrovytsky,, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here:
My research:
X/Twitter:
Thumbnail design: Felícia Zsolnai-Fehér –
Bro speaking in token per second
Needs the cerebras chip upgrade 😂
It’s as if there is a full stop after each word.
😂😂😂
Straight out of the from the Uncanny valley. if this is not AI, I am even more confused.
I am an old subscriber and i like his content but I really wish he take some narration lessons 😂
I was just reading their threads, sounds super interesting.
Plus is a max of 5 seconds (it appears to be 5 seconds of 720p or 10 seconds of 480p) and pro is a max of 20 seconds
Plus users ($20 monthly) will be able to generate 50 clips a month. Pro users ($200 monthly) will be able to generate 500 clips + unlimited on a slower speed.
WHAT A TIME TO BE ALIVE! ( 4:48 QUEUE DANCING SUSHI EMOTE )😃
I just say : Hunyuan Video from tencent – local !
The problem with text-to-video is that 80% of what you generate is garbage. When you try to generate something slightly complex, it fails miserably.
The amount of energy wasted on those things…
And it’s well known fact in any visual media that getting in to 80-90% is the easy part. That last 10-20% usually takes up the majority of time. It might be that AI video models (with current approach) can’t break through that 90% barrier. Image quality has improved massively but same underlying issues are very much still there.
Which might be ok if you could do 100 iterations to get the result you want. But the token cost is prohibitive even if you are paying £200 a month.
Don’t look at where we are now, but where we will be two papers down the line.
@@elwendyrofys5944you could tell them a hundred times and you will still see the same comment
A good usecase for these models is prototyping VFX. AI is good at imagining whimsical effects that can then be recreated in artist digital creation tools. Otherwise using Sora (or the other models) on its own can’t cut it.
The incoherence has a dreamlike quality to it.
Yeah it’s almost human
Goodbye reality. Nothing on screen can be deemed evidentially real anymore.
Yeah I wish we could just go back to the prehistoric era when everything you saw online could be trusted!
We still can make the difference most of the time. But in few years (months?) it’s wouldn’t be possible
OpenAi while having most of the ai not open source is ironic
At this point I have very few interest in closed source ai.
or you could just say its fraud
You mean the nonprofit organisation that has a “capped profit arm” wich rakes in millions using the training data they were only allowed to use as a non profit research? :O
2:41 I appreciate the “hold on to your papers” on-screen visual alert.
“We will all be able to become film directors”. Nightmare fuel.
The way that AI messes up is too similar to our dreams
My thoughts exactly. Results from AI image/video generation do seem very similar to how our dreams look.
“A stroke of the brush does not guarantee art from the bristles.” – Kosh
🖖🤬👍
I took too long time to release… And still, not released on my country (Brazil).
When it was announced I was extremely impressed, but so much time has passed, so many competitors, that it’s not THAT impressive anymore.
I’m much more hyped by hunyuan
when we all become able to be film directors, there will be no film directors.
I still feel like I entered an alternate timeline in 2022 with how AI is so rapidly advancing when all we had was nightmare fuel pictures 2,5 years ago. Looking forward to local models!
5:20 Damn that seems quite optimistic
with unreal5 and 2 friends we can make AAA games in 2 weeks. now, with this, we can make holywood-quality AAA movies with 0 friends in 2 days.