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

SANA-WM, a 2.6B open-source world model for 1-minute 720p video

nvlabs.github.io

Post reply on HN
Print view View original

141–150 of 162 posts

Re: SANA-WM, a 2.6B open-source world model for 1-minute 720p video

#141

Post by stephen_cagle »

dasyatidprime wrote: post #128
stephen_cagle wrote: post #120

Earlier quoted context omitted.

I feel like a circular saw is a creator tool, while a microwave is mostly a finishing tool. I take raw material and make something out of it with a circular saw, largely unrestrained by anything other than cost, skill, and material. With a microwave, I make things hot so I can eat them. Aside: Also, I wonder why that is? Why do we regard the microwave as "degenerate" compared to the oven? Why is baking seen as a call…

You might both be interested in https://malmesbury.substack.com/p/my-journey-to-the-microwav... regarding the history of microwave cooking (not mine). The microwave oven wasn't known destined to be relegated to lowbrow cookery from the beginning, even if that's how it turned out from our perspective, and some of the more advanced techniques developed for it fell out of use.

Neat article.

I think it is interesting (though I only partially agree) that microwave meals require standardization to scale. Let's say that was true, why couldn't a modern microwave have a small camera and a set of heuristics for how to cook just about anything by turning the gun on and off at particular points when it recognizes a food? Maybe without intelligence, a microwave does need standardization; but we can put intelligence (ideally offline) in just about anything these days?

I wonder if with sufficient control if a microwave could ever brown? I wonder if it could reliably bake?

https://www.youtube.com/watch?v=2tg3-93jKvc - Chicken Good!

Re: SANA-WM, a 2.6B open-source world model for 1-minute 720p video

#142

Post by altmanaltman »

mccoyb wrote: post #16

I struggle with these world models from the perspective of video games (so this post is a particular perspective). I'm not a game developer myself, but some of my favorite games carry a deep sense of intentionality. For instance, there is typically not a single item misplaced in a FromSoftware game (or, for instance, Lies of P -- more recently). Almost every object is placed intentionally. Games which lack this inten…

Re what are we building towards - I think ultimately AI and these tools will be very useful to professionals who already know how to excel at their art.

They can be very powerful tools but creating meaningful art/subjective work that is actually good is actually a borderline impossible task for LLMs since it requires genuine creativity.

Similarly, I am seeing coding agents be infinitely more useful to people who already code than non-technical people thinking they can vibe code their own Salesforce.

Its just that AI doesn't work well as a "productivity booster" tool in the marketing context. So they are pushing towards the idea that anyone can do anything with AI which is imo a really stupid hill to die on or base your ipo on.

Re: SANA-WM, a 2.6B open-source world model for 1-minute 720p video

#143

Post by Schiendelman »

DrewADesign wrote: post #136
afzalive wrote: post #129

Earlier quoted context omitted.

Just like with mass-produced materials vs hand-crafted stuff, you're gonna have a lot of crap quality and rare, expensive good quality stuff.

And most people can’t just spin up a furniture factory at their whim and call themselves a designer. AI gives everybody with the slightest gumption a fully-functional, “initially plausible crap” factory at their fingertips, so everybody with actual skills gets lost in a sea of useless garbage.

Discovery is always the problem to solve - discovering the good products is hard.

Re: SANA-WM, a 2.6B open-source world model for 1-minute 720p video

#144

Post by hatsix »

stirfish wrote: post #125
hattmall wrote: post #123

Earlier quoted context omitted.

Tbh, most people actually don't know how to microwave. The typical microwave users just trys to "one-shot" it by punching in a time and using high heat. Proper microwaving is what gave rise to the entire concept of "fast casual" restaurants, famously AppleBees (or "club B's" in the late night focus iterations!) Complex entrees that could be partially cooked and frozen. Then rapidly microwaved on a custom program that…

I have a collection of vintage microwave cookbooks and they get real fancy, with techniques like wrapping the thin parts of fish with aluminum foil so they don't burn. Volume 5 had a full Thanksgiving dinner. I think the goal was to sell more microwaves to people who weren't sure what they could do. Fascinating stuff. I have used none of those recipes. The microwave is for making cold pizza 10% more palatable (or 80%…

You reheat pizza in the microwave? I hope someone cooks salmon in your microwave!

Re: SANA-WM, a 2.6B open-source world model for 1-minute 720p video

#145

Post by WithinReason »

mccoyb wrote: post #16

I struggle with these world models from the perspective of video games (so this post is a particular perspective). I'm not a game developer myself, but some of my favorite games carry a deep sense of intentionality. For instance, there is typically not a single item misplaced in a FromSoftware game (or, for instance, Lies of P -- more recently). Almost every object is placed intentionally. Games which lack this inten…

To turn this into an actual usable game the network would need to generate an image conditionally from (low asset quality) maps made by the game devs, so map design, object placement is human, graphics is AI.

Re: SANA-WM, a 2.6B open-source world model for 1-minute 720p video

#146

Post by DrewADesign »

Schiendelman wrote: post #143
DrewADesign wrote: post #136

Earlier quoted context omitted.

And most people can’t just spin up a furniture factory at their whim and call themselves a designer. AI gives everybody with the slightest gumption a fully-functional, “initially plausible crap” factory at their fingertips, so everybody with actual skills gets lost in a sea of useless garbage.

Discovery is always the problem to solve - discovering the good products is hard.

And when there’s an ocean of bullshit, it’s a lot harder, which makes the incentive to do good things nearly zero. Scale is very important, here.

Re: SANA-WM, a 2.6B open-source world model for 1-minute 720p video

#148

Post by dasyatidprime »

stephen_cagle wrote: post #141
dasyatidprime wrote: post #128

Earlier quoted context omitted.

You might both be interested in https://malmesbury.substack.com/p/my-journey-to-the-microwav... regarding the history of microwave cooking (not mine). The microwave oven wasn't known destined to be relegated to lowbrow cookery from the beginning, even if that's how it turned out from our perspective, and some of the more advanced techniques developed for it fell out of use.

Neat article. I think it is interesting (though I only partially agree) that microwave meals require standardization to scale. Let's say that was true, why couldn't a modern microwave have a small camera and a set of heuristics for how to cook just about anything by turning the gun on and off at particular points when it recognizes a food? Maybe without intelligence, a microwave does need standardization; but we can…

Most of the problem with browning is the physics: it requires temperatures well above boiling, but the default microwave absorber is water which will tend to saturate around boiling temp. But physical cookware design can be used to make it work: the article I linked prominently describes the use of a browning skillet with a tin oxide coating on the bottom that can safely absorb large amounts of microwave radiation and get hot enough to brown foods.

So partly due to that, I think if you wanted to do full robotic cookery with a microwave beyond the TV-dinner stage you'd still need control over the other objects rather than just switching the magnetron. Stirring for even heat distribution is another thing.

And then it's separately interesting how, in a really skew way, “deliver heat using RF energy” came back from a different angle with the rise of induction stoves!

Re: SANA-WM, a 2.6B open-source world model for 1-minute 720p video

#149

Post by derangedHorse »

DrewADesign wrote: post #146
Schiendelman wrote: post #143

Earlier quoted context omitted.

Discovery is always the problem to solve - discovering the good products is hard.

And when there’s an ocean of bullshit, it’s a lot harder, which makes the incentive to do good things nearly zero. Scale is very important, here.

Luckily the same AI tools that are generating the content can be used to build better tailored discriminators for it as well. If I can define what it is I dislike about an essay, video, etc., and give it to an LLM, it can tell me whether a piece I present to it is worthwhile to consume according to my standards. This even applies to things the LLM can't generate things for that meet the same standard for its programmed/prompted discrimination.

Re: SANA-WM, a 2.6B open-source world model for 1-minute 720p video

#150

Post by DrewADesign »

stephen_cagle wrote: post #140
DrewADesign wrote: post #137

Earlier quoted context omitted.

And theoretically AI does a great job at helping HR filter unqualified candidates, and it helps candidates optimize their resumes and application strategies to help them land the right role. So people should be landing dream roles left-and-right. Is that how it’s working? In reality, I don’t see any of this trending towards the theoretical happy path everybody always talks about. Most people give up trying to find so…

Just to be clear, I'm visualizing the usage of world models that can consistently render visual and interactive renderings of a specified world. I think interacting with them will be markedly different than interacting with many text based LLMs (though I don't know, I have never had direct access to one). I don't think these will create "artist" in any sense, but I do think it will lower the barrier dramatically for…

Sorry I probably conflated LLMs and diffusion models, et al in what I wrote, but there’s no real difference in art.

Artists make thousands of tiny decisions when creating art— many of them unconscious. It’s true across mediums— e.g. digital art, oil painting, collage, sculpture, and architecture photography— and even realms, like music, visual art, and animation. Learning to make art is a process of learning to make those conscious and unconscious decisions to create something that communicates what you want it to, using the subtle visual language developed by everything from your personality, to your physical capability and limitations, your cognitive capability and limitations, your experience, taste, likes, available tools and media, environment, inspiration, etc. etc. etc.

AI simultaneously makes it harder to make those decisions and imbue your work with your artistic perspective while supplanting it with the amalgamated decisions of other artists. Novices love AI because they don’t have to learn how to make those artistic judgements, or the craft of implementing them… unfortunately, learning that stuff is learning art. Using AI to generate very specific images can be a difficult, deep, and creative process, but it is fundamentally different than creating art— it is art direction. The skills to do it are fundamentally different. It’s not an incremental change like physical media to digital media, which requires the same exact underlying set of cognitive skills and processes even if the logistics were different.

You could definitely consider a meaningfully curated collection of images other people (or machines) generated to be ‘art,’ but no matter how closely an art director micromanaged an artist making an individual piece, it is still, fundamentally, the artist’s low-level sensibility, experience, and perspective on the micro level that made that piece what it was. If that art director put their name on that artist’s piece, they’d justifiably be run out of town with pitchforks.

But AI does not make learning art more accessible to people. They make it nearly free to commission art from a computer. This isn’t a purely philosophical distinction — it would be like someone saying they’re an experienced driver because they use Waymo cars. Even if they get a new ride every block to make sure they take the exact route they want to take, that doesn’t mean they’re driving. Even if you could tell the Waymo to speed up, slow down, or change lanes, you’re still back-seat driving at best. It’s just a different thing. And if you’re banking on more Waymo cars making it easier for people to get in cars, and increasing the number of car trips to increase the number of experienced drivers in the world, that’s a fundamentally flawed way of reasoning about it.

Post reply on HN