It's a big launch, but AI video-synthesis competition has heated up over the past 10 months.
See full article...
See full article...
Ten months after first tease, OpenAI launches Sora video generation publicly
- Thread starter JournalBot
- Start date
Looks great, but after waiting so long, I really expected it to be a step above the competition—and it's not. (I guess they focused on making it faster rather than better.)
The competition is just as good, less censored, and offers features not available from Sora.
There are also some open-source models that recently came out and seem just as capable as this one (but you need some monster hardware to run them.)
I was kind of hoping OAI would be the first to roll out a model that could generate both audio and video, but I guess we'll have to wait until next year for that.
The competition is just as good, less censored, and offers features not available from Sora.
There are also some open-source models that recently came out and seem just as capable as this one (but you need some monster hardware to run them.)
I was kind of hoping OAI would be the first to roll out a model that could generate both audio and video, but I guess we'll have to wait until next year for that.
Upvote
30
(32
/
-2)
Looks like their sign in servers are down from the browser and Sora is not available in iOS app yet.
Looking forward to playing with Sora. Still ambivalent about what down stream effects Sora will entail; AI slop, misinformation, etc.
Looking forward to playing with Sora. Still ambivalent about what down stream effects Sora will entail; AI slop, misinformation, etc.
Upvote
2
(4
/
-2)
How long will the tech coast along on boasting the quality of everything they CAN realistically incrementally improve, before inevitably crashing into the hard walls of longterm temporal consistency, in combination with finegrained detailed adjustments?
How close are we to the "95% there!"?
How close are we to the "95% there!"?
Upvote
55
(56
/
-1)
Artem S. Tashkinov
Ars Scholae Palatinae
Marques Brownlee (I'm not subscribed to him but he was the first to post an extensive review) has reviewed Sora:
View: https://www.youtube.com/watch?v=OY2x0TyKzIQ
It might be useful for experimenting with creative ideas and very short very specific clips, but when it comes to motion, especially involving real world physics or limbs (for some reason), it struggles hard. It also struggles with continuity and hallucinates a lot.
And as far as I understood, it's silent at the moment, i.e. there's no audio generation.
Don't expect it to fake reality just yet. Maybe in a year or two.
View: https://www.youtube.com/watch?v=OY2x0TyKzIQ
It might be useful for experimenting with creative ideas and very short very specific clips, but when it comes to motion, especially involving real world physics or limbs (for some reason), it struggles hard. It also struggles with continuity and hallucinates a lot.
And as far as I understood, it's silent at the moment, i.e. there's no audio generation.
Don't expect it to fake reality just yet. Maybe in a year or two.
Last edited:
Upvote
17
(22
/
-5)
The music video felt a bit blah, in that the generated transitions just felt random. It makes me think that this is going to require a human drafting specific prompts very granularly to get anything compelling out of AI.
Upvote
13
(13
/
0)
Considering these are the clips and prompts hand-picked by OpenAI, I am very underwhelmed.
The soap dragon doesn't really "breathe" bubbles the way a dragon breathes fire.
The hero-dog loops like a poorly composited greenscreen from a 1990s pet store advertisement. The background looks very generic and not like NYC at all. (Except if you look closely, there are four (4) buildings that kinda-sorta look like the Empire State Building). It's actually worse upon repeated viewings because the tail kind of disappears and reappears from under the cape in an unnatural manner that makes me think the AI lacks object permanence.
The soap dragon doesn't really "breathe" bubbles the way a dragon breathes fire.
The hero-dog loops like a poorly composited greenscreen from a 1990s pet store advertisement. The background looks very generic and not like NYC at all. (Except if you look closely, there are four (4) buildings that kinda-sorta look like the Empire State Building). It's actually worse upon repeated viewings because the tail kind of disappears and reappears from under the cape in an unnatural manner that makes me think the AI lacks object permanence.
Upvote
66
(66
/
0)
For now...2 years, 5 years, 10 years down the road and we are in something completely different. There is no putting the genie back in the bottle.
Upvote
-10
(9
/
-19)
Just as I expected, OpenAI hyped Sora months ago, then vanished. Now that they have actually put something out, the reviews are not great. (High censored and struggling with image consistency seem to be the big complaints.) Meanwhile group like Kling and Minimax are offering more, for less.
Will OpenAI continue to be where the investment goes, or will people with piles of money to burn start looking elsewhere? There is an investment to be made here. Even Sora, with its shortcomings, is going to be a formidable tool in the hands of skilled human user. I just think OpenAI, perhaps due to the hubris of Altman, is slow walking themselves out of being competitive.
95%? Hell 90% is more than enough to be considered highy successful .
Will OpenAI continue to be where the investment goes, or will people with piles of money to burn start looking elsewhere? There is an investment to be made here. Even Sora, with its shortcomings, is going to be a formidable tool in the hands of skilled human user. I just think OpenAI, perhaps due to the hubris of Altman, is slow walking themselves out of being competitive.
95%? Hell 90% is more than enough to be considered highy successful .
Upvote
-1
(6
/
-7)
Regarding the hero image for this article, why do mushroom-headed creatures have breasts? They must be going for the fans of Jennifer Aniston from Friends.
Upvote
0
(8
/
-8)
A good use of energy? Given generating one still image uses a similar amount of energy to a full phone battery.
Upvote
0
(9
/
-9)
The AI was definitely getting confused over whether it was a tail or just an extra red flap beyond the blue cape. Also, the length of the tail varied from one unobscured appearance to the next.
Upvote
4
(4
/
0)
Soulless text > soulless pictures > soulless videos
Burning up the planet for shitty knockoffs to save a few bucks.
Burning up the planet for shitty knockoffs to save a few bucks.
Upvote
55
(64
/
-9)
Wow, that music video was too busy.. I think the guy needed to slim it down a bit. Less is more, right?
Upvote
5
(5
/
0)
mushroom people crowd mushroom woman in foreground mushroom beige stock footage poppin nips 4K UHD these nips are blastin beautiful sexy mushroom woman HDR full beams if you know what i mean trending on artstation trending on youtube high quality jennifer aniston type thing going on with the nipples photorealistic uncanny surreal directed by jj abrams wonderful
Upvote
68
(71
/
-3)
"Mushroom nipples" was not on the list of things I expected to see today.
But it makes sense; porn-ification of all the things is far and away the leading use case for AI.
Upvote
41
(42
/
-1)
It must be hard to train a model to come up with 10-20s videos when anything filmed this century has a cut at most every 7 seconds, and generally a lot less than that.
Upvote
22
(22
/
0)
Nah, it is just a tool, and it will be the creative people who are just learning the tool and how to push it that will put the soul back into it. It is the same evolution that occurred when the Macintosh landed on every graphic designer's desk.
Upvote
-10
(7
/
-17)
The "playing" part should probably keep OpenAI executives awake at night.
Sure, they get their sub fees as people play, but once the curiosity is sated, what's left? A cancelled account because it's just not useful to, well, anyone for anything that's not directly related to an income or for things that will soon be, if not already is, illegal.
And I'd think that the vast majority of netizens have no shits to give about AI video generation if they have something HONEST to present.
That leaves the cheaters, propagandists, liars, haters, and other folks who pollute the gene pool for them to rely on for income. One wonders if that's a viable business model.
Upvote
5
(6
/
-1)
The heads of those mushrooms all appear similar, as does the folds in their clothing. Demo was impressive but this is meh.
Upvote
0
(0
/
0)
Who do you think will use this instead of hiring people to make professional video? If you're using this it's because you either can't afford to, or simply will not pay someone to do it (e.g.: maybe I might to want a video of a dachshund juggling for funsies, I'm sure not paying someone hundreds of dollars to do it).
A band using generative AI to make a music video would have, instead, shot it themselves with a camera phone.
Upvote
-3
(4
/
-7)
"Just a tool" sure is doing a lot of heavy lifting here. Calling something "the same evolution" merely serves to obfuscate the important differences between the introduction of the Mac and the introduction of AI.
Upvote
8
(10
/
-2)
Upvote
10
(10
/
0)
90% is fine if you have an expert reviewing the output and there’s little cost to generating new answers. But that expert costs a lot.
If you want something a non-expert can pick up and use, it needs to be a product, a tool - it needs to just work. And a tool that fails 1 in 10 times is a terrible tool. If a chatbot gives confidently wrong answers to 1 in 10 questions, and is fielding 1000s of requests, that’s hundreds of complaints the company now has to address.
There’s a reason eBay sellers tout 99% positive reviews, and companies aim for six sigma.
Edit: another issue with calling GenAI tools “tools” Is that a tool does what its user intends. GenAI is a poor tool by that standard, both by the 1 in 10 failures and in the difficulty in making fine adjustments.
Last edited:
Upvote
29
(29
/
0)
$20 vs $200.
10x the price for 200x the value.
This New-AI world will further cement the planet's riches and resources into the hands of the few; the wealthy.
10x the price for 200x the value.
This New-AI world will further cement the planet's riches and resources into the hands of the few; the wealthy.
Upvote
0
(3
/
-3)
Are you seriously telling me that those clips - the "adorable" foam creature that looks like poorly animated frog spawn, the "looks like cheaply produced 80's TV" Superdog clip with background snow that looks utterly wrong because the tool doesn't understand scale/perspective - those clips are the best that OpenAI could produce to showcase this new tech?
Bad job OpenAI. Bad. Job.
I continue to be amazed at how easily impressed some people are.
As a final note, how much of a wanker do you have to be to shart out a <2min AI-generated music video and then refer to it as a "film".
Fuck all the way off.
Bad job OpenAI. Bad. Job.
I continue to be amazed at how easily impressed some people are.
As a final note, how much of a wanker do you have to be to shart out a <2min AI-generated music video and then refer to it as a "film".
Fuck all the way off.
Upvote
23
(27
/
-4)
Why is the article's top image a still from someone's mushroom porn fantasy? I wasn't even going to click on the article but the associated image was so weird I had to comment on it.
Upvote
13
(14
/
-1)
If the image wasn’t specifically cherry-picked (though it probably was), this could be a side effect of efforts to expand gender and race representation in the models. We’ve seen this recently with improbable races in historical photos, emoji genders in iOS 18.2, etc.
As the training data and reinforcement encourages a wider range, the outputs become more likely to depict combinations we might find to be unusual to our reality-based sensibilities.
I can’t imagine how much harder that makes it to prevent showing us results that are downright racist or pornographic, being generated basically via roll of the dice.
Upvote
7
(8
/
-1)
There are two dimensions of "make sense" here; the first is about how the model works. The second (and to me more interesting) is "of the functionally infinitely many things to show, why did the meatbags choose that?"
Upvote
2
(2
/
0)
Thing is: it IS all mighty impressive in a tech demo kinda sense
But the extrapolation that is necessary to further lift it up to what it is imagined to be in the future, doesn't seem to have much to start from. Something entirely novel will have to be injected to get there. Which is entirely possible, I guess, but WHAT currently actually warrants any supreme confidence in that? It all just feels far too similar to the self driving cars story.
It's like they managed to train a human weightlifter to suddenly lift twice as heavy weights than anyone ever did before, and people would be sold the idea (and believe) that, soon, human construction workers will replace cranes altogether.
Upvote
7
(7
/
0)