Agreed. Everyone raves about generative AI but I've yet to see a single generative AI video that is "enjoyable" to watch in a way beyond the way tech demos are enjoyable to watch.
That was also true of quite a lot of early CGI, but most people would say that things have improved. I think we're on the cusp of rapid improvement in AI video as well, in part spurred on by skilled people using the tools we currently have.
I came across the following recently. I think a casual viewer would assume it was just Bakshi-style rotoscoped animation without a major AI component.
My constructive criticism to this video is "90% of it is figures standing still while wind blows their outfit or the camera does a simple move." Sometimes moving their lips as though talking .. though I did like that bird turning it's head smoothly away like "forget this, I'mma preen! Peace out!" Haha
Very much no throughline of concepts from one shot to the next. You never see the same character twice. No foreground dynamic action.. not even simple walking except one far-away character directly away from the camera which means that their silhouette hardly changed.
This all comes from the current generation of video diffusion models that basically just generate an image like they always have except with a hint of temporal coherence they expand that into a short shot with no types of movement except those seen a million times in their training set.
Getting gen models to be able to reason better about motion and to build mental world models of the 3d scene they are managing a 2d window into is going to be a big challenge, and require some additional breakthroughs on a par with the original GPT and stable diffusion breakthroughs that currently act as a foundation to a majority of modern AI innovation.
> ... and require some additional breakthroughs on a par with the original GPT and stable diffusion breakthroughs ...
You say this like Stable Diffusion isn't a 2022 technology. And not early 2022, but quite late (August). ChatGPT is younger.
I mean sure we need more breakthroughs, but we've barely even seen a new hardware generation since those things came out and the researchers are really only getting started with the new capabilities of generative tech. If we don't get more breakthroughs in short order then that would be a stunning halt of progress, a breaking stop the likes of which we have almost never before seen. More breakthroughs are a given.
An interesting example. It may be because I consider myself a fan of animation (moreso than the average person), but the video has obvious garbage less than fifteen seconds in, with the spaceships (?) morphing and sludging around the pyramid.
Sure, that's why I said casual viewer and not careful viewer. But getting back to your original point, would you say it was foul and unpleasant? That's really what I'm claiming, that we're fairly quickly advancing beyond the old days of those nightmare Nekobuses and vomit-inducing clips of Will Smith devouring spaghetti, and into territory where at least some people can find the product genuinely enjoyable. Of course nothing will ever be perfect. AI aside, after all these years it's still often jarring when computer physics is shoehorned into cartoons/anime that's designed to look like traditional hand drawn animation.
If I were to watch 90+ minutes of that with dubbed voices on top of it, absolutely. There's practically zero cohesion between any of those shots. No real action, no real narrative. It's a collection of non-cohesive stills that were stretched, not any bit of a story at all.
Not sure why you're downvoted. This is one of the most objectively true things said here. CGI was pretty crappy for at least the first few decades of its existence. Even aspects of the animation in Toy Story really show that film's age. I remember realizing that in the early 2000's. Most people either forgot or didn't even experience the early days of CGI and would consider much of it to be nightmare fuel today.
AI is pretty clearly advancing orders of magnitude faster than CGI has. Just because it sucks now doesn't mean it's going to suck in another 5 years.
"Just because it sucks now doesn't mean it's going to suck in another 5 years."
We will see. Some flaws might be baked in, like LLM's halucinating. That won't go away, unless we invent a new tech.
So here with generating videos, will morphing objects for example ever go away? I am sceptical with the current approach.
To me the magic is in generating things that would have had too much right issues or would not have been economically viable.
A series that is a variant of the stories of “The Wire”, but taking place in the Harry Potter universe? Coming right up.
Obscure prog rock band from the 90s put out one album? Now its two.
I can understand people their apprehension, feeling like art is losing something essential without the human touch behind it, but I saw an article a few days ago where people thought generated Shakespeare was better than actual Shakespeare. Until it was revealed which was generated.
If AI can generate me another, better Illmatic, I’m all here for it.
The conclusion to draw there is that Shakespeare honestly doesn't make for very good reading today unless you are also delving into the historical context or considering the major impact of his works on humanity's culture from the arts to language. LLM generated output has none of that.
'Ah ha! LLMs are better than Shakespeare!' is a meaningless statement.
Besides, no one reads Shakespeare for pleasure; there is no need to generate more. ;)
I bet you could get a random person off the street to fairly-consistently pick (curated) AI works over an amalgamated top-5 of great jazz recordings selected by jazz super-fans.
> Besides, no one reads Shakespeare for pleasure
Exactly (kind of). Lots of rewarding works take effort to learn to appreciate, for a bunch of reasons that may include (as in Shakespeare's case) that they're old and their context and vernacular is not ours. Lots of people (I'd say a large majority, in my experience) dismiss entire genres and forms of art that they weren't heavily exposed to as children, often going so far as to judge them bad, simply because it would take some time and effort to learn how to enjoy them and to be able to discern what's good or remarkable about a given work.
What proportion of the population has ever in their lives enjoyed the experience of reading a Shakespeare play? Even once? It's gotta be tiny. Of course you can get them to choose AI junk over Shakespeare, it's not a kind of thing they understood or enjoyed to begin with, in most cases.
> I know the work I'm doing is valuable and that this field is the future. I'm sure it'll click for more folks soon.
Maybe because not everyone shares your opinion? Having an LLM generate art isn't necessarily a net benefit for society. Computers were supposed to improve our lives but instead of robots to perform dangerous menial work it's taking the creativity out of humanity.
Hey guys you no longer have to do fun things, tech bros have that covered. Now get back down the mine.
The thing is GenAI will be just like CGI. When it is bad it will look bad and has 'that look'. But when it is good enough you will not even know.
The creativity has already gone sideways for most of this. I can with a few simple sentences create an acceptable picture (in some cases a short film). With a AI pipeline I can make some pretty cool scenes. Instead of having to know how to properly draw an s curve with a nice gradient bit of layered colors over it and 14 meticulously created layers. I tell the program to do it for me. It does an acceptable job in a fraction of the time. People can complain all they want but the rest of us are already using these tools and will continue to do so until something better comes along.
Most of the people who seem to be fans of AI-generated art are fans of AI, not of art.
Maybe it'll get to the point where it's good enough to have on as background television - not everything needs to be great, after all - but what's the point of that? We already have far more high-quality television shows and movies than most people can ever watch.
So far, it's really bad at actually replacing human labor in pro-social ways while being a supercharger for various antisocial jobs, like scam artist or astroturfer. The main "beneficial" use for it today is replacing wasteful labor that probably didn't need to be done in the first place—which is why an AI version is fine, because it didn't matter to begin with.
My wife and I both work in the field, I on the tech side, her on the creative side, and she's been in it since the earliest days of industry trying to adapt these tools. There's a lot (like, holy shit, so much) of effort and money going into it, but so far it's only marginally helpful for non-evil jobs.
I was speaking in response to most of the general "pervasive negative attitudes" mentioned, not you specifically. Although I'm curious where you viewed generative AI content repeatedly 5 years ago; it was effectively non-existent outside research circles then.
If I recall correctly, that was about the time that Google started demoing its generative AI "deepmind" - a particular demo of a frog comes to mind. The commonality of AI content has certainly increased since then, I didn't mean to imply it was commonplace back then.
People might care for culture. I’m very much in favor of a reformed copyright that strengthens indie artists, conservationists, and remixers and weakens Disney et al.
It’s also a matter of fact that we have the copyright we have that’s prohibitive to people and favors corporations. It’s upsetting to see how a bunch of Silicon Valley companies stomps right across those lines with impunity, while people like Aaron Swartz are persecuted and threatened with decade long prison sentences for crimes that in my mind ought to be much less upsetting.
If copyright was fair, training of AI intended for non-personal use ought to be a sufficient commercial activity to require a license. That would stiffle the development of AI, which is what I’d argue happens to human creators under our current system.
If we had a 25 year copyright, we could easily make useful AI trained on the sum of human creation until 1999, _and_ have badass human made remixes of 80s and 90s songs — we wouldn’t have to do legal gymnastics to allow the development of useful AI, as it’d have access to quite substantial training material from the 1900s, and unlock relatively modern training material year-by-year.
So yes, I dislike AI for infringing on copyright and I dislike copyright (in its current state).
Disagree. I've seen things done with generative AI which I've wanted to see visualized for decades, but which were too difficult and time-consuming to do traditionally. I've also seen beautiful things that were either impossible to produce or impossible to even conceive of through traditional workflows.
Those make up an infinitesimally small portion of the total output, which is largely a deluge of crap, certainly. But, generally, rarity makes something more valuable and beautiful by comparison.
It is not inherently, but people are very effective at using it to produce foul, unpleasant output, which is a temporary problem. Like almost all things, people will not actually care how they're made if the final product is good.
