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

Veo

deepmind.google

Post reply on HN
Print view View original

361–370 of 539 posts

Re: Veo

#361

Post by sdenton4 »

salamo wrote: post #353

The first thing I will do when I get access to this is ask it to generate a realistic chess board. I have never gotten a decent looking chessboard with any image generator that doesn't have deformed pieces, the correct number of squares, squares properly in a checkerboard pattern, pieces placed in the correct position, board oriented properly (white on the right!) and not an otherwise illegal position. It seems to be…

This strikes me as equally "AI complete" as drawing hands, which is now essentially a solved problem... No one test is sufficient, because you can add enough training data to address it.

Re: Veo

#362

Post by arcticbull »

pmlarocque wrote: post #360
arcticbull wrote: post #354

Earlier quoted context omitted.

Similarly the Veo example of the northern lights is a really interesting one. That's not what the northern lights look like to the naked eye - they're actually pretty grey. The really bright greens and even the reds really only come out when you take a photo of them with a camera. Of course the model couldn't know that because, well, it only gets trained on photos. Gets really existential - simulacra energy - maybe a…

That not true, they look grey when they aren't bright enough, but they can look green or red to the naked eyes if they are bright. I have seen it myself and yes I was disappointed to see only grey ones last week. see: https://theconversation.com/what-causes-the-different-colour...

> [Aurora] only appear to us in shades of gray because the light is too faint to be sensed by our color-detecting cone cells."

> Thus, the human eye primarily views the Northern Lights in faint colors and shades of gray and white. DSLR camera sensors don't have that limitation. Couple that fact with the long exposure times and high ISO settings of modern cameras and it becomes clear that the camera sensor has a much higher dynamic range of vision in the dark than people do.

https://www.space.com/23707-only-photos-reveal-aurora-true-c...

This aligns with my experiences.

The brightest ones I saw in Northern Canada I even saw hints of reds - but no real greens - until I looked at it through my phone, and it looked just like the simulated video.

If I looked up and saw them the way they appear in the simulation, in real life, I'd run for a pair of leaded undies.

Re: Veo

#363

Post by SJC_Hacker »

Eji1700 wrote: post #338
ugh123 wrote: post #219

From a filmmaking standpoint I still don't think this is impactful. For that it needs a "director" to say: "turn the horse's head 90˚ the other way, trot 20 feet, and dismount the rider" and "give me additional camera angles" of the same scene. Otherwise this is mostly b-roll content. I'm sure this is coming.

There's also the whole "oh you have no actual model/rigging/lighting/set to manipulate" for detail work issue. That said, I personally think the solution will not be coming that soon, but at the same time, we'll be seeing a LOT more content that can be done using current tools, even if that means a dip in quality (severely) due to the cost it might save.

This lead me to the question of why hasn't there been an effort to do this with 3D content (that I know of).

Because camera angles/lighting/collision detection/etc. at that point would be almost trivial.

I guess with the "2D only" approach that is based on actual, acquired video you get way more impressive shots.

But the obvious application is for games. Content generation in the form of modeling and animation is actually one the biggest cost centers for most studios these days.

Re: Veo

#364

Post by jprete »

candiddevmike wrote: post #345
xnx wrote: post #212

60 second example video: https://www.youtube.com/watch?v=diqmZs1aD1g

For some reason this video reminds me of dreaming--details just kind of pop in and out and the entire thing seems very surreal and fractal.

Same impression here. The scene changes very abruptly from a sky view to following the car. The cars meld with the ground frequently, and I think I saw one car drive through another at one point.

Re: Veo

#365

Post by Tronno »

arcticbull wrote: post #362
pmlarocque wrote: post #360

Earlier quoted context omitted.

That not true, they look grey when they aren't bright enough, but they can look green or red to the naked eyes if they are bright. I have seen it myself and yes I was disappointed to see only grey ones last week. see: https://theconversation.com/what-causes-the-different-colour...

> [Aurora] only appear to us in shades of gray because the light is too faint to be sensed by our color-detecting cone cells." > Thus, the human eye primarily views the Northern Lights in faint colors and shades of gray and white. DSLR camera sensors don't have that limitation. Couple that fact with the long exposure times and high ISO settings of modern cameras and it becomes clear that the camera sensor has a much…

I've seen it bright green with the naked eye. It definitely happens. That article is inaccurate.

Re: Veo

#366

Post by salamo »

sdenton4 wrote: post #361
salamo wrote: post #353

The first thing I will do when I get access to this is ask it to generate a realistic chess board. I have never gotten a decent looking chessboard with any image generator that doesn't have deformed pieces, the correct number of squares, squares properly in a checkerboard pattern, pieces placed in the correct position, board oriented properly (white on the right!) and not an otherwise illegal position. It seems to be…

This strikes me as equally "AI complete" as drawing hands, which is now essentially a solved problem... No one test is sufficient, because you can add enough training data to address it.

Yeah "AI complete" is a bit tongue-in-cheek but it is a fairly spectacular failure mode of every model I've tried.

Re: Veo

#367

Post by shermantanktop »

sdenton4 wrote: post #359
arcticbull wrote: post #354

Earlier quoted context omitted.

Similarly the Veo example of the northern lights is a really interesting one. That's not what the northern lights look like to the naked eye - they're actually pretty grey. The really bright greens and even the reds really only come out when you take a photo of them with a camera. Of course the model couldn't know that because, well, it only gets trained on photos. Gets really existential - simulacra energy - maybe a…

That doesn't seem in any way useful, though... To use a very blunt analogy, are color blind people intelligent/sentient/whatever? Obviously, yes: differences in perceptual apparatus aren't useful indicators of intelligence.

As a colorblind person…I could see the northern lights way better than all the full-color-vision people around me squinting at their phones.

Wider bandwidth isn’t always better.

Re: Veo

#368

Post by Jonanin »

TaylorAlexander wrote: post #125
TIPSIO wrote: post #115

Earlier quoted context omitted.

Objectively speaking (if people would be honest with themselves), both are just decent at best. I think comparing them now is probably not that useful outside of this AI hype train. Like comparing two children. A lot can happen. The bigger message I am getting from this is it's clear OpenAI won't have a super AI monopoly.

Comparing two children is a good one. My girlfriend has taken to pointing out when I’m engaging in “punditry”. They're an engineer like I am and we talk about tech all the time, but sometimes I talk about which company is beating which company like it’s a football game, and they call me out for it. Video models are interesting, and to some extent trying to imagine which company is gonna eat the other’s lunch is kind…

Except that many of the people involved do think of it like a football game, and thus it actually is like one. Of course the researchers and engineers at both OpenAI and Google DeepMind have a sense of rivalry and strive to one up another. They definitely feel like they are in a competition.

Re: Veo

#370

Post by paxys »

arcticbull wrote: post #354
salamo wrote: post #353

The first thing I will do when I get access to this is ask it to generate a realistic chess board. I have never gotten a decent looking chessboard with any image generator that doesn't have deformed pieces, the correct number of squares, squares properly in a checkerboard pattern, pieces placed in the correct position, board oriented properly (white on the right!) and not an otherwise illegal position. It seems to be…

Similarly the Veo example of the northern lights is a really interesting one. That's not what the northern lights look like to the naked eye - they're actually pretty grey. The really bright greens and even the reds really only come out when you take a photo of them with a camera. Of course the model couldn't know that because, well, it only gets trained on photos. Gets really existential - simulacra energy - maybe a…

That's not true at all. I have seen northern lights with my own eyes that were more neon green and bright purple than any mainstream photo.
Post reply on HN