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

Meta Movie Gen

ai.meta.com

Post reply on HN
Print view View original

491–500 of 1001 posts

Re: Meta Movie Gen

#491

Post by jetrink »

itslennysfault wrote: post #479
jetrink wrote: post #450

Earlier quoted context omitted.

A friend who lives in North Carolina sent me a video of the raging floodwaters in his state- at least that's what the superimposed text claimed it was. When I looked closer, it was clearly an Indian city filled with Indian people and Indian cars. He hadn't noticed anything except the flood water. It reminded me of that famous selective attention test video[1]. I won't ruin it for those who haven't seen it, but it's a…

hmmm... Maybe it's because I knew it was testing me, but I noticed it right away and counted the right count. I could see it being pretty shocking if I hadn't, but I honestly can't imagine how I'd miss that.

It probably doesn't work if you're primed to look for hidden details. I took the test along with my Psychology 101 class of about 30 people and no one noticed anything amiss.

Re: Meta Movie Gen

#492

Post by CSSer »

hackernewds wrote: post #489
jetrink wrote: post #450

Earlier quoted context omitted.

A friend who lives in North Carolina sent me a video of the raging floodwaters in his state- at least that's what the superimposed text claimed it was. When I looked closer, it was clearly an Indian city filled with Indian people and Indian cars. He hadn't noticed anything except the flood water. It reminded me of that famous selective attention test video[1]. I won't ruin it for those who haven't seen it, but it's a…

I do not see how the examples you mentioned are related to the topic? What does selective attention have to do with the video looking AI generated in all the frames?

Their argument is that if someone is affected by confirmation bias, they likely won’t notice these kinds of details.

Essentially, send me a video of something I care about and I will only look for that thing. Most people are not detectives, and even most would-be detectives aren’t yet experts.

Re: Meta Movie Gen

#493

Post by burkaman »

psb217 wrote: post #409
burkaman wrote: post #304

Earlier quoted context omitted.

Yes, and like pretty much every AI release I've seen, even these cherry-picked examples mostly do not quite match the given prompt. The outputs are genuinely incredible, but if you imagine actually trying to use this for work, it would be very frustrating. A few examples from this page: Pumpkin patch - Not sitting on the grass, not wearing a scarf, no rows of pumpkins the way most people would imagine. Sloth - that's…

The one thing that immediately stood out to me in the ghost example was how the face of the ghost had "wobbly geometry" and didn't appear physically coupled to the sheet. This and the way the fruit in the sloth's drink magically rested on top of the drink without being wedged onto the edge of the glass as that would require were actually some of the more immediate "this isn't real" moments for me.

I think those types of visual glitches can probably be fixed with more or better training, and I have no doubt that future versions of this type of system will produce outputs that are indistinguishable from real videos.

But better training can't fix the more general problem that I'm describing. Perfect-looking videos aren't useful if you can't get it to follow your instructions.

Re: Meta Movie Gen

#494

Post by williamcotton »

dinfinity wrote: post #442
benabbott wrote: post #395

Things are about to get weird. We can't control this at any level: At the level of image/video synthesis: Some leading companies have suggested they put watermarks in the content they create. Nice thought, but open source will always be an option, and people will always be able to build un-watermarked tools. At the level of law: You could attempt to pass a law banning image/video generation entirely, or those without…

The challenge is to determine what is real, not what is fake. I think cryptographic signing and the classic web of trust approaches are going to prove the most valuable tools in doing so, even if they're definitely not a panacea.

The web of trust doesn't seem to scale! All of the online social platforms trend towards centralization for identify verification.

In my (historically unpopular) opinion we have two optional choices outside of but still allowing for this anonymous free-for-all:

A private company like Facebook uses a privileged system of identification and authentication based on login/password/2FA and relying on state-issued identification verification,

OR, what I feel is better, a public institution that uses a common system based on PKI and state-issued identification, eg, the DMV issuing DoD Common Access Cards.

Trusting districts and nation-states could sign each other's issuing authorities.

The benefits are multifaceted! It helps authenticate the source of deep fakes. It helps fight astroturfing, foreign or otherwise. It helps to remove private companies fueled by advertising revenue from being in a privileged position of identification, etc, etc.

I totally understand any downvotes but I would prefer if you instead engaged me in this conversation if you disagree.

I'd love to have this picked apart instead of just feeling bummed out.

Re: Meta Movie Gen

#495

Post by paul7986 »

echoangle wrote: post #418
paul7986 wrote: post #378

Earlier quoted context omitted.

I agree with Neil Degrasse Tyson AI will kill the Internet https://www.youtube.com/watch?v=SAuDmBYwLq4 Though maybe there's hope if.. 1. All deepfake image & video tech are enforced to add watermark labels & all websites that publish are force to label fake too. 2. Crazy idea but a govt issued Internet ID (ID.me is closest to that now with having to use to file taxes with the IRS) where your personal repuation and cr…

1. will never happen, it’s way too interesting that people won’t try to make an open source version where the watermark can easily be removed by users. Unless you actually criminalize it and put people in jail multiple years for building anything close to deepfakes, you won’t be able to prevent that.

That's why websites uploaders need to read the meta data and found out how it was originally generated then publish / label it ..AI generated it its first creation was such.

Re: Meta Movie Gen

#496

Post by blargey »

schmidtleonard wrote: post #230
beezlebroxxxxxx wrote: post #198

Earlier quoted context omitted.

I'll be blunt, all of those images look comically generic and extremely "AI". > Photography, digital painting, 3D rendering Those are not the same as AI. Using AI is akin to standing beside a great pianist and whispering into his ear that you want "something sad and slow" and then waiting for him to play your request. You might continue to give him prompts but you're just doing that. In time, you might be called a "c…

> I'll be blunt, all of those images look comically generic and extremely "AI". Ok, now I know you're watching through hate goggles. Fortunately, not everyone will bring those to the party. > Using AI is akin... [goes on to describe a clueless iterative prompting process that wouldn't get within a mile of the front page] You've really outed yourself here. If you think it's all just iterative prompting, you are about…

I scrolled through and...have to agree with their impression. I'm confused as to what you thought is being demonstrated by images on https://civitai.com/images of all things, since it's all very high-concept/low-intentionality, to put it nicely. Did you mix it up with a different link?

Re: Meta Movie Gen

#497

Post by vunderba »

jcims wrote: post #103

One thing I've noticed with the set of music generation tools (eg Udio, Suno) is that there's a sort of profound attachment to songs that you create. I've never made music the old fashioned way so I'm guessing the same could be true for that as well, but there are songs I've made on Udio that I personally think are amazing but nobody else really responds to. Conversely I can see similar levels of pride and attachment…

One thing I've noticed with the set of music generation tools (eg Udio, Suno) is that there's a sort of profound attachment to songs that you create.

With all due respect, how could there be when at the click of a button you can generate entire songs? You didn't come up with the chord progression, the structure, the melodic motifs, or the lyrics.

My attachment to my works is directly proportional to the amount of effort it took to create them.

Re: Meta Movie Gen

#498

Post by yunwal »

firebaze wrote: post #487
itslennysfault wrote: post #479

Earlier quoted context omitted.

hmmm... Maybe it's because I knew it was testing me, but I noticed it right away and counted the right count. I could see it being pretty shocking if I hadn't, but I honestly can't imagine how I'd miss that.

> hmmm... Maybe it's because I knew it was testing me, but I noticed it right away and counted the right count. > I could see it being pretty shocking if I hadn't, but I honestly can't imagine how I'd miss that. The point of the video wasn't to count correctly, but to see the gorilla

cool, he noticed it right away

Re: Meta Movie Gen

#499

Post by derefr »

Twisell wrote: post #445
throwup238 wrote: post #335

Earlier quoted context omitted.

A surprising amount of it is really popular too. I recently figured out that the Movie Recaps channel was all AI when the generated voice slipped and mispronounced a word in a really unnatural way. They post videos almost daily and they get millions of views. Must be making bank.

Most probably they employ overseas, underpaid workers with non-standard English accents and so they include text-to-speach in the production process to smoothen the end result. I won't argue wether text to speech qualifies as an AI but I agree they must be making bank.

> Most probably they employ overseas, underpaid workers with non-standard English accents and so they include text-to-speach in the production process to smoothen the end result.

Might also be an AI voice-changer (i.e. speech2speech) model.

These models are most well-known for being used to create "if [famous singer] performed [famous song not by them]" covers — you sing the song yourself, then run your recording through the model to convert the recording into an equivalent performance in the singer's voice; and then you composite that onto a vocal-less version of the track.

But you can just as well use such a model to have overseas workers read a script, and then convert that recording into an "equivalent performance" in a fluent English speaker's voice.

Such models just slip up when they hit input phonemes they can't quite understand the meaning of.

(If you were setting this up for your own personal use, you could fine-tune the speech2speech model like a translation model, so it understands how your specific accent should map to the target. [I.e., take a bunch of known sample outputs, and create paired inputs by recording your own performances of them.] This wouldn't be tenable for a big low-cost operation, of course, as the recordings would come from temp workers all over the world with high churn.)

Re: Meta Movie Gen

#500

Post by kylebenzle »

CSSer wrote: post #485
dekhn wrote: post #451

Earlier quoted context omitted.

That sheen looks (to me) like some of the filters that are used by people who copy videos from TV and movie and post them on (for example) facebook reels. There's an entire pattern of reels that are basically just ripped-off-content with enough added noise to (I presume) avoid content detection filters. Then the comments have links to scam sites (but are labelled as "the IMDB page for this content").

The idea that Meta’s effectively stolen content is tainted by a requirement to avoid collecting stolen content is laughably ironic.

Yes, but thats just a hypothesis, have we seen any evidence that shows the cause of the "AI sheen" is bad training data, or more likly, just a shortcomming of generating realistic photos from text at this early stage.
Post reply on HN