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

151–160 of 539 posts

Re: Veo

#151

Post by Jensson »

fakedang wrote: post #63
loudmax wrote: post #21

The videos in this demo are pretty neat. If this had been announced just four months ago we'd all be very impressed by the capabilities. The problem is that these video clips are very unimpressive compared to the Sora demonstration which came out three months ago. If this demo was announced by some scrappy startup it would be worth taking note. Coming from Google, the inventor of the Transformer and owner of the larg…

Honestly, if Veo becomes public faster than Sora, they could win the video AI race. But what am I wishfully thinking - it's Google we're talking about!

> But what am I wishfully thinking - it's Google we're talking about!

Google the company known to launch way too many products? What other big company launches more stuff early than them? What people complain about Google is that they launch too much and then shut them down, not that they don't launch things.

Re: Veo

#152

Post by BlackJack »

airstrike wrote: post #39

> Veo > Sign up to try VisionFX Is it Veo or VisionFX? Is it a sign up, a trial, or a waitlist? How hard can it be to write a clear message? In the words of Don Miller, if you confuse, you lose.

Disclaimer: I work at Google on related stuff

Veo is the name of a video model. VideoFX is the name of a new experimental tool at labs.google.com, which uses Veo and lets you make videos.

Thanks for the feedback though, I see how it's confusing for users.

Re: Veo

#153

Post by SoftTalker »

Vaguely unsettling that the thumbnail for first example prompt "A lone cowboy rides his horse across an open plain at beautiful sunset, soft light, warm colors" looks something like the pixelated vision of The Gunslinger android (Yul Brynner's character) from the 1973 version of Westworld.

See 1:11 in this video https://www.youtube.com/watch?v=MAvid5fzWnY

Incidentally that was one of the early uses of computer graphics in a movie, supposedly those short scenes took many hours to render and had to be done three times to achieve a colorized image.

Re: Veo

#154

Post by lupire »

Jensson wrote: post #101
popcar2 wrote: post #46

Not nearly as impressive as Sora. Sora was impressive because the clips were long and had lots of rapid movement since video models tend to fall apart when the movement isn't easy to predict. By comparison, the shots here are only a few seconds long and almost all look like slow motion or slow panning shots cherrypicked because they don't have that much movement. Compare that to Sora's videos of people walking in rea…

> Sora was impressive because the clips were long and had lots of rapid movement Sora videos ran at 1 beat per second, so everything in the image moved at the same beat and often too slow or too fast to keep the pace. It is very obvious when you inspect the images and notice that there are keyframes at every whole second mark and everything on the screen suddenly goes in their next animation step. That really limits…

So it needs to learn how far each object can travel in 1sec at its natural speed?

Re: Veo

#155

Post by mike_hearn »

airstrike wrote: post #39

> Veo > Sign up to try VisionFX Is it Veo or VisionFX? Is it a sign up, a trial, or a waitlist? How hard can it be to write a clear message? In the words of Don Miller, if you confuse, you lose.

Presumably this is DeepMind vs Labs fighting over the same project. A consequence of guaranteeing Demis some level of independence when DeepMind was bought, which still shows through in the fact that the DeepMind brand(s) survive.

Re: Veo

#156

Post by AceJohnny2 »

SoftTalker wrote: post #153

Vaguely unsettling that the thumbnail for first example prompt "A lone cowboy rides his horse across an open plain at beautiful sunset, soft light, warm colors" looks something like the pixelated vision of The Gunslinger android (Yul Brynner's character) from the 1973 version of Westworld. See 1:11 in this video https://www.youtube.com/watch?v=MAvid5fzWnY Incidentally that was one of the early uses of computer graphi…

Can't say I see a visual similarity. In any case, "Cowboy silhouette in the sunset" is a pretty classic American visual.

But the parallel you made between android Brynner's vision and the generated imagery is fun to consider!

Re: Veo

#157

Post by kvetching »

testfrequency wrote: post #145
typpo wrote: post #114

The amount of negativity in these comments is astounding. Congrats to the teams at Google on what they have built, and hoping for more competition and progress in this space.

[flagged]

hatred fuels our capitalism more than anything

Re: Veo

#158

Post by redml »

axblount wrote: post #48

I hate to be so cynical, but I'm dreading the inevitable flood of AI generated video spam. We really are about this close to infinite jest. Imagine TikTok's algorithm with on demand video generation to suit your exact tastes. It may erase the social aspect, but for many users I doubt that would matter too much. "Lurking" into oblivion.

I think of it as we're replacing the SEO spam we have right now with AI spam. At least now we can fight that with more AI.

Re: Veo

#159

Post by lupire »

kristofferR wrote: post #150
ryandrake wrote: post #77

Earlier quoted context omitted.

Obligatory: Liam Neeson jumps over a fence in 6 seconds, with 14 cuts[1]. 1: https://www.youtube.com/watch?v=gCKhktcbfQM

The top comment makes a really good point though: "He's 68. I'm guessing they stitched it together like this because "geriatric spends 30 seconds scaling chainlink fence then breaks a hip" doesn't exactly make for riveting action flick fare." Lingering shots are horrible for obscuring things.

Movies have stunt performers.

And Neeson was only 60 when filming Taken 3.

Re: Veo

#160

Post by Jensson »

lupire wrote: post #154
Jensson wrote: post #101

Earlier quoted context omitted.

> Sora was impressive because the clips were long and had lots of rapid movement Sora videos ran at 1 beat per second, so everything in the image moved at the same beat and often too slow or too fast to keep the pace. It is very obvious when you inspect the images and notice that there are keyframes at every whole second mark and everything on the screen suddenly goes in their next animation step. That really limits…

So it needs to learn how far each object can travel in 1sec at its natural speed?

It also needs to separate animation steps for different objects so that objects can keep different speeds. It isn't trivial at all to go from having a keyframe for the whole picture to having separate for separate parts, you need to retrain the whole thing from the ground up and the results will be way worse until you figure out a way to train that.

My point is that it isn't obvious at all that Soras way actually is closer to the end goal, it might look better today to have those 1 second beats for every video but where do you go from there?

Post reply on HN