Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Sora 2

openai.com

Post reply on HN
Print view View original

651–660 of 916 posts

Re: Sora 2

#651

Post by rixed »

raincole wrote: post #560
wrigby wrote: post #553

Earlier quoted context omitted.

There are two completely distinct differences that jump out to me initially that I think may help justify your feelings: 1: Reading a long book demands focus on a longer timespan than scrolling TikTok, and with focusing on a single thing for a long time, we get a sense of accomplishment. I don’t know how to justify this as valuable, but for some reason I feel that it is. 2: The Song of Ice and Fire (and GoT) were con…

I think the point 2 will rub many people the wrong way (me included) though. That would make reading Fourth Wing or Twilight a more connecting experience than most classics. (Nothing inherently wrong with that, but... you know...)

Depends if you are trying to connect to your contemporaries or to mankind in general. Aren't "classics" just timeless pop?

Re: Sora 2

#652

Post by Klonoar »

minimaxir wrote: post #587

This Sora 2 generation of Cyberpunk 2077 gameplay managed to reproduce it extremely closely , which is baffling: https://x.com/elder_plinius/status/1973124528680345871 > How the FUCK does Sora 2 have such a perfect memory of this Cyberpunk side mission that it knows the map location, biome/terrain, vehicle design, voices, and even the name of the gang you're fighting for, all without being prompted for any of those s…

> Everyone already assumed that Sora was trained on YouTube

Doesn't this already answer your question...? "Let's Play" type videos and streams have been a thing for years now, even for more obscure games. It very well could've been trained on Cyberpunk videos of that mission.

Re: Sora 2

#653

Post by vunderba »

School-Cotton wrote: post #531
ascorbic wrote: post #397

Earlier quoted context omitted.

There isn't a standard written form of any major sign languages

Yes, but there’s no fundamental reason why there couldn’t be one. It’s not a good reason to accept all the downsides of AI.

AI is irrelevant to the reason why there isn't a written version of every single national dialect of sign language. The reason it doesn't exist is because it would serve no purpose (source: many deaf friends). Deaf communities learn the country's writing system just like everyone else.

The closest thing out there is SignWriting [1] which has about as much traction in the real world as esperanto.

https://en.wikipedia.org/wiki/SignWriting

Re: Sora 2

#654

Post by minimaxir »

Klonoar wrote: post #652
minimaxir wrote: post #587

This Sora 2 generation of Cyberpunk 2077 gameplay managed to reproduce it extremely closely , which is baffling: https://x.com/elder_plinius/status/1973124528680345871 > How the FUCK does Sora 2 have such a perfect memory of this Cyberpunk side mission that it knows the map location, biome/terrain, vehicle design, voices, and even the name of the gang you're fighting for, all without being prompted for any of those s…

> Everyone already assumed that Sora was trained on YouTube Doesn't this already answer your question...? "Let's Play" type videos and streams have been a thing for years now, even for more obscure games. It very well could've been trained on Cyberpunk videos of that mission.

It's hard for me to believe that the model coherently memorized both the video and audio of a relatively obscure Let's Play, and that a simple prompt was enough to surface it (the use of the term "Basilisk tank" would also likely not be in video metadata either). That is the reason the person who made that tweet, who has far more prompting experience than myself, was shocked.

Re: Sora 2

#655

Post by tootie »

gorgoiler wrote: post #448
tootie wrote: post #442

Earlier quoted context omitted.

The fact that this is their demo to the world and it's full of errors implies that average users will only get worse results.

I’m wary of being that damning, this early. What I want to know is, should my video have these kinds of continuity errors, how easily can I fix them? It’s ok for this to be a fun toy. (And fun toy while also being an astonishing piece of engineering.) But if it wants to push beyond fun toy then it would be interesting to see how that process works. Will Sora2 help me sketch out a movie for me, doing 10% of the work w…

I don't see how this is usable for making like a feature film. Editing will be impossible. At best it will be for ads. At worst for making social media slop.

Re: Sora 2

#656

Post by non_sequitur »

bsenftner wrote: post #364
qgin wrote: post #169

VFX artists are definitely feeling the AGI / considering other career paths today.

VFX artist and developer here, who's deep into this stuff, and it is really not there. It's an island of itself, barely controllable and barely usable with other media. They are just now getting around to generating alpha channels, with virtual none of the existing pipelines for any AI video or image generation tools to even incorporate and work with alpha channels. This is just one of several hundred aspects of inco…

Honest q - do you think these things will make a big difference if these videos can be made in 15 minutes for $20 or whatever?

Won’t the industry change to adopt that massive price cut/productivity gain?

Re: Sora 2

#657

Post by echelon »

crucialfelix wrote: post #414
the_duke wrote: post #273

I haven't seen comments regarding a big factor here: It seems like OpenAI is trying to turn Sora into a social network - TikTok but AI. The webapp is heavily geared towards consumption, with a feed as the entry point, liking and commenting for posts, and user profiles having a prominent role. The creation aspect seems about as important as on Instagram, TikTok etc - easily available, but not the primary focus. Genera…

> It seems like OpenAI is trying to turn Sora into a social network - TikTok but AI. That's a direct copy of what Midjourney has done already. https://www.midjourney.com/explore?tab=videos Many people are just playing with images and the distinctive styles that Midjourney (the model) seems to have developed. It's also trained by ratings and people's interactions. When you make images you can dial down the "aesthetic"…

After seeing the results of Sora 2, Meta and Midjourney are not yet competing at this level.

This is the "Suno" moment for video.

It's easy to make a really compelling composition. Something even Google Veo couldn't do.

It's not the best looking video model, but it has everything else -- rich editing, good voices and lipsync, music and lyrics, animation (cartoon, 3D, anime), SFX. It's wild.

The videos aren't single clips but rather complete beginning-middle-end stories that unfold over several cuts.

Re: Sora 2

#658

Post by bsenftner »

gmueckl wrote: post #449
bsenftner wrote: post #364

Earlier quoted context omitted.

VFX artist and developer here, who's deep into this stuff, and it is really not there. It's an island of itself, barely controllable and barely usable with other media. They are just now getting around to generating alpha channels, with virtual none of the existing pipelines for any AI video or image generation tools to even incorporate and work with alpha channels. This is just one of several hundred aspects of inco…

Do you even see a path from the current AI systems to something that has that near-total control over every detail that is required for high quality VFX work?

Yes. Adding alpha channels would be step one. Then perhaps incorporate the "element" concept that is basically any identifiable visual anything; which is what VFX uses as a composite-capable element. Then build a whole visual scene description prose that is what we give to a video AI, and that prose is high level language where necessary and element-wise specific where necessary. Base that scene description prose on the language used by film makers directly, just adopt their terminology, and then track the industry's jargon within the models. That way anyone working in media will auto-magically know how to control them.

We are at a point now where it is now how to write software that is the problem but how to describe to the software that is the problem. Video and film making is so generalized, AI needs more information. Typically that information comes from a director's and their team's consistency during production. AI has neither the information for consistency of imagery nor the narrative and the perspective of the narrative a human director and team bring. In time, AI will develop large enough contexts, but will the hardware to run that be affordable? There is a huge amount of context in both an entire script and the world view perspective a film crew brings to any script, and for that reason I think many of the traditional (VFX included) film roles are not going to suddenly disappear. AI video does not replace their consistency at their budget, hands down.

When AI video is able to be just a part of the skill set, for example when it is compatible with compositing, editing, and knows that terminology, AI video will be adopted more. Right now, it is designed as an all or nothing offering.

Re: Sora 2

#659

Post by q3k »

varispeed wrote: post #297
qoez wrote: post #27

I know the comments here are gonna be negative but I just find this so sick and awesome. Feels like it's finally close to the potential we knew was possible a few years ago. Feels like a pixar moment when CG tech showed a new realm of what was possible with toy story

I still feel this is limited by what it learned from. It looks cool but it also looks like something I'd dreamt or saw flicking through TV channels. Kind of like spam for the eyes.

It looks like it has been trained exclusively on car advertisement videos playing at airports.

Re: Sora 2

#660

Post by bsenftner »

non_sequitur wrote: post #656
bsenftner wrote: post #364

Earlier quoted context omitted.

VFX artist and developer here, who's deep into this stuff, and it is really not there. It's an island of itself, barely controllable and barely usable with other media. They are just now getting around to generating alpha channels, with virtual none of the existing pipelines for any AI video or image generation tools to even incorporate and work with alpha channels. This is just one of several hundred aspects of inco…

Honest q - do you think these things will make a big difference if these videos can be made in 15 minutes for $20 or whatever? Won’t the industry change to adopt that massive price cut/productivity gain?

The cost is and will be more than that, the time will be more, and I really think people are underestimating the time it takes to create good stories. Sure, there will be online locations to make short form video of all kinds. People have had video cameras in their pockets for a very long time and being hobby film makers are not really popular. The AI video sites now are 95% people fascinated with the ability to make video at all, and after a bit their interest dies because to actually make anything that requires real work even with AI helping left and right. Consistency is a harsh mistress; and AI video is only good with it for a short duration. So any narrative that makes a story worth watching, it's not AI slop, will continue to require humans and human creativity - for the consistency that gives a story the integrity that makes it worth watching. At least for audiences that care. No doubt, there are commercial forces working to develop audiences that like and prefer AI slop.
Post reply on HN