Live data from Hacker News

Meta Movie Gen

ai.meta.com

621–630 of 1001 posts

Re: Meta Movie Gen

#621
post #504

Earlier quoted context omitted.

The subtle "errors" are all low hanging fruit. It reminds me of going to SIGGRAPH years back and realizing most of the presentations were covering things which were almost imperceptible when looking at the slides in front. The math and the tech was impressively, but qualitatively it might have not even mattered. The only interesting questions now have nothing to do with capability but with economics and raw resources…

> In a few years, or less, clearly we'll be able to take our favorite books and watch unabridged, word-for-word copies. The quality, acting, and cinematography will rival the biggest budget Hollywood films. The "special effects" won't look remotely CG like all of the newest Disney/Marvel movies -- unless you want them to. If publishers put up some sort of legal firewall to prevent it, their authors, characters, and s…

Indeed, adverts come first.

But at the same time, while it is indeed true that the end result is far more than simply just making good images, LLMs are weird interns at everything — with all the negative that implies as well as the positive, so they're not likely to produce genuinely award winning content all by themselves even though they can do better by asking them for something "award winning" — so it's certainly conceivable that we'll see AI indeed do all these things competently at some point.

Re: Meta Movie Gen

#622

Earlier quoted context omitted.

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.

This comes up a lot. Because synthesis is so generally feasible plus the existence of very powerful editing tools for things like movies and whatnot, I'm guessing that it will simply become the norm to assume that any image, sound, movie, or whatever may be fake. I expect there won't be a way to verify something was synthesised or "real-synthesized" (since images and videos are ultimately synthesized themselves, just…

Trusted entities could vouch for the veracity (or other aspects) of things, especially if they are close to the source.

We already implicitly do this: if a news outlet we trust publishes a photo and does not state that they are unsure of its veracity we assume that it is an authentic photo. Using cryptographic signing that news outlet could explicitly state that they have determined the photo to be real. They could add any type of signed statement to any bit of information, really. Even signing something as being fake could be done, with the resulting signed information being shareable (although one would imagine that any unsigned information would be extremely suspect anyway).

The web of trust approach is to have a distributed system of trust that allows for less institutional parties to be able to earn trust and provide 'trusted' information, but there are also plenty downsides to it. A similar distributed system that determines trustworthiness in a more robust way would be preferable, but I am not aware of one.

Re: Meta Movie Gen

#623

Earlier quoted context omitted.

There are usable ones runwayml.com pika.art hailuoai.com

klingai.com lumalabs.ai

When I see lists of URLs like that I can only wonder what a future post archeologist, coming upon this long dusty thread half a decade from now, will find when they try to go to those sites.

Re: Meta Movie Gen

#624
post #281

> Upload an image of yourself and transform it > into a personalized video. Movie Gen’s > cutting-edge model lets you create personalized > videos that preserve human identity and motion. A stalker’s dream! I’m sure my ex is going to love all the videos I’m going to make of her! Jokes aside, it’s a little bizarre to me that they treat identity preservation as a feature while competitors treat that as a bug, explicitl…

Your dream $11.99 kitchen knife! Perfect for stabbing! IT IS REAL!

* product made without use of AI or any unnatural components. pure mountain iron

Re: Meta Movie Gen

#625
post #281

> Upload an image of yourself and transform it > into a personalized video. Movie Gen’s > cutting-edge model lets you create personalized > videos that preserve human identity and motion. A stalker’s dream! I’m sure my ex is going to love all the videos I’m going to make of her! Jokes aside, it’s a little bizarre to me that they treat identity preservation as a feature while competitors treat that as a bug, explicitl…

One positive aspect of it is that at some point people will just not care about nudes. Which is better for the victims of rageporn, not worse.

Re: Meta Movie Gen

#626

All the vids have that instantly recognizable GenAI "sheen", for the lack of a better word. Also, I think the most obvious giveaway are all the micro-variations that happen along the edges, which give a fuzzy artifact.

I did some images generation and found a LORA for VHS footage. It's amazing what "taking away the sheen" can do to make an image look strikingly real.

Re: Meta Movie Gen

#627

So I’m probably going to be too closed minded about this: but who the f*ck asked for this and did anyone consider consequences of easily accessible AI slop generation? It’s already nearly impossible to find quality content on the internet if you don’t know where to look at.

I did and I'm quite happy that this is happening :) It's unleashing a new computing era when you just have to lean back, close your eyes and your vision can materialize without a Hollywood production crew.

And it's great as anyone can use it in whichever way they want since machine generated content does not have copyright protections.

We will finally achieve the dream of everything being in public domain!

Re: Meta Movie Gen

#628
post #409

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.

The ghost is insanely impressive, it's the example that gave me a "wow" effect. The cloth physic looks stunning, I never thought we would reach such a level of temporal coherence so fast.

Re: Meta Movie Gen

#629

Why do these video generation ones never become usable to the public. Is it just they had to create millions of videos and cherry pick only a handful of decent generations? Or is it just so expensive there's no business model for it? My mind instantly assumes it a money thing and they're just wanting to charge millions for it, therefore out of reach for the general public. But then with Meta's whole stance on open ai…

There are a few available to the public. runway.ai and kling are a couple that I see heavily used on Twitter. I pay for runway right now for experiments and it works. The problem is that maybe 1 out of 10 prompts result in something useable. And when I say useable I have pretty low standards. Since the model pumps out 5 or 10 second clips you have to be pretty creative since the models still struggle with keeping any…

Hi zoogeny (and anyone else here) — you can try our new app Nim to address the Runway problems you describe https://alpha.nim.video

We offer both image-to-video (same situation as Runway, need a few attempts to make something awesome) and video-to-video (under the name "Restyle 2.0") - this is our newest tool and is highly reliable, i.e. you can get complex motion (kissing, handshakes, boxing, skateboarding, etc) with controllable changes to input video (changing outfits, characters, backgrounds, styles).

Unlike Runway and Kling, we currently offer a smiple UNLIMITED plan for just $10/mo. Check it out! https://alpha.nim.video

Re: Meta Movie Gen

#630

Earlier quoted context omitted.

It's still very different. What you describe is exactly what an art director does, which is creative and difficult— there's a good reason many commercial artists end their careers as art directors but none start there. Anybody that says making things that look good and interesting using generative AI is easy or doesn't require genuine creativity is just being a naysayer. However, at most, the art director is credited…

> your still working with other people's borrowed humanity usefully pureed and reformed by a machine Exactly, isn't it amazing? You can travel the latent space of human culture in any direction. It's an endless mirror house where you can explore. I find it an inspiring experience, it's like a microscope that allows zooming into anything.

Sure it's a lot of fun. I also find it very useful for some things like references and mood boards. No matter how granular you get with control nets or LORAs and how good the models get, you just can't get the specificity needed for professional work and the forms it gives you are just too onerous to mold into a useful shape using professional tools. It's still, fundamentally, asking another thing to make it for you, like work for hire or a commission. Software like Nuke's copycat tool or Adobe's background remover or content-aware fill were professionally useful right off the bat because they were designed for professional use cases. Even then, text prompt image generators are more useful than not in low-effort, high-volume use cases where the extremely granular per-pixel nuance doesn't really matter. I doubt they'll ever be useful enough for anything higher-level than that. It's just fundamentally the wrong interface for this work. It's like saying a bus driver on a specific route with a bus is equally useful to a cab driver with a cab. There are obviously instances where that's true, but no matter how many great things you can show are on that bus route, and no matter how many people it's perfectly suited for, there's just no way a FedEx driver could use it to replace their van.
Post reply on HN