Apple Releases Open Weights Video Model • Hacker News Board

Re: Apple Releases Open Weights Video Model

#111

Post by shagie » Tue, Dec 02, 2025, 4:30 PM UTC

andy_ppp wrote: post #104

devinprater wrote: post #6

Apple has a video understanding model too. I can't wait to find out what accessibility stuff they'll do with the models. As a blind person, AI has changed my life.

I wonder if there's anything that can help blind people to navigate the world more easily - I guess in the future AR Glasses won't just be for the sighted but allow people without vision to be helped considerably. It really is both amazing and terrifying the future we're heading towards.

From a couple years ago...

https://www.microsoft.com/en-us/garage/wall-of-fame/seeing-a...

https://youtu.be/R2mC-NUAmMk

https://youtu.be/DybczED-GKE

... and that was 10 years ago. I'm curious for what it could do now.

Re: Apple Releases Open Weights Video Model

#112

Post by tippa123 » Tue, Dec 02, 2025, 4:32 PM UTC

joedevon wrote: post #21

tippa123 wrote: post #19

Earlier quoted context omitted.

+1 and I would be curious to read and learn more about it.

If you want to see more on this topic, check out (google) the podcast I co-host called Accessibility and Gen. AI.

Honestly, that’s such a great example of how to share what you do on the interwebs. Right timing, helpful and on topic. Since I’ve listened to several episodes of the podcast, I can confirm it definitely delivers.

Re: Apple Releases Open Weights Video Model

#113

Post by shagie » Tue, Dec 02, 2025, 4:38 PM UTC

dzhiurgis wrote: post #31

gostsamo wrote: post #27

Earlier quoted context omitted.

Not the gp, but currently reading a web novel with a card game where the author didn't include alt text in the card images. I contacted them about it and they started, but in the meantime ai was a big help. all kinds of other images on the internet as well when they are significant to understanding the surrounding text. better search experience when Google, DDG, and the like make finding answers difficult. I might us…

As a (web app) developer I never quite sure what to put in alt. Figured you might have some advice here?

I'm gonna flip this around... have you tried pasting the image (and the relevant paragraph of text) and asking ChatGPT (or another LLM) to generate the alt text for the image and see what it produces?

For example... https://chatgpt.com/share/692f1578-2bcc-8011-ac8f-a57f2ab6a7...

Re: Apple Releases Open Weights Video Model

#114

Post by vjvjvjvjghv » Tue, Dec 02, 2025, 4:45 PM UTC

MyFirstSass wrote: post #102

Workaccount2 wrote: post #93

Earlier quoted context omitted.

People need to understand that a lot of angst around AI comes from AI enabling people to do things that they formally needed to go through gatekeepers for. The angst is coming from the gatekeepers. AI has been a boon for me and my non-tech job. I can pump out bespoke apps all day without having to get bent on $5000/yr/usr engineering software packages. I have a website for my side business that looks and functions pr…

I highly doubt "pumping out bespoke apps all day" is possible yet besides 100% boilerplate, and when possible then no good for any other purpose than enshittifiying the web, and at that point not profitable because everyone can do it. I use AI daily as a senior coder for search and docs, and when used for prototyping you still need to be a senior coder to go from say 60% boilerplate to 100% finished app/site/whatever…

This is the same as the discussion about using Excel. Excel has its limitations, but it has enabled millions of people to do pretty sophisticated stuff without the help of “professionals”. Most of the stuff us tech people do is also basically some repetitive boilerplate. We just like to make things more complex than they need to be. I am always a little baffled why seemingly every little CRUD site that has at most 100 users needs to be run on Kubernetes with several microservices, CI/CD pipelines, and whatever.

As far as enshittification goes, this was happening long before AI. It probably started with SEO and just kept going from there.

Re: Apple Releases Open Weights Video Model

#115

Post by dragonwriter » Tue, Dec 02, 2025, 4:48 PM UTC

embedding-shape wrote: post #86

andersa wrote: post #82

The number of video models that are worse than Wan 2.2 and can safely be ignored has increased by 1.

To be fair, the sizes aren't comparable, and for the variant that is comparable, the results aren't that much worse.

The samples (and this may or may not be completely fair, either set could be more cherry picked than the other, It would be interesting to see a side-by-side comparison with comparable prompts) seem significantly worse than what I’ve seen from WAN 2.1 1.3B, which is both fron the previous WAN version and is smaller, proportionally, compared to Apple’s 7B than that model itself is compared to the 28B combination of the high and low noise 14B WAN 2.2 models that are typically used together.

But also, Starflow-V is a research model with a substandard text encoder, it doesn't have to be competitive as-is to be an interesting spur for further research on the new architecture it presents. (Though it would be nice if it had some aspect where it offered a clear improvement.)

Re: Apple Releases Open Weights Video Model

#116

Post by cubefox » Tue, Dec 02, 2025, 4:52 PM UTC

Interesting that this is an autoregressive ("causal") model rather than a diffusion model.

Re: Apple Releases Open Weights Video Model

#117

Post by doug_durham » Tue, Dec 02, 2025, 4:57 PM UTC

embedding-shape wrote: post #83

GeekyBear wrote: post #58

Earlier quoted context omitted.

One cool feature they added for deaf parents a few years ago was a notification when it detects a baby crying.

Is that something you actually need AI for though? A device with a sound sensor and something that shines/vibrate a remote device when it detects sound above some threshold would be cheaper, faster detection, more reliable, easier to maintain, and more.

You are talking about a device of smart phone complexity. You need enough compute power to run a model that can distinguish noises. You need a TCP/IP stack and a wireless radio to communicate the information. At that point you have a smart phone. A simple sound threshold device would have too many false positives/negatives to be useful.

Re: Apple Releases Open Weights Video Model

#118

Post by Invictus0 » Tue, Dec 02, 2025, 5:04 PM UTC

Apple's got to stop running their AI group like a university lab. Get some actual products going that we can all use--you know, with a proper fucking web UI and a backend.

Re: Apple Releases Open Weights Video Model

#119

Post by Rover222 » Tue, Dec 02, 2025, 5:10 PM UTC

densh wrote: post #7

devinprater wrote: post #6

Apple has a video understanding model too. I can't wait to find out what accessibility stuff they'll do with the models. As a blind person, AI has changed my life.

> As a blind person, AI has changed my life. Something one doesn't see in news headlines. Happy to see this comment.

`Something one doesn't see` - no pun intended

Re: Apple Releases Open Weights Video Model

#120

Post by Aurornis » Tue, Dec 02, 2025, 5:30 PM UTC

embedding-shape wrote: post #83

GeekyBear wrote: post #58

Earlier quoted context omitted.

One cool feature they added for deaf parents a few years ago was a notification when it detects a baby crying.

Is that something you actually need AI for though? A device with a sound sensor and something that shines/vibrate a remote device when it detects sound above some threshold would be cheaper, faster detection, more reliable, easier to maintain, and more.

> more reliable

I've worked on some audio/video alert systems. Basic threshold detectors produce a lot of false positives. It's common for parents to put white noise machines in the room to help the baby sleep. When you have a noise generating machine in the same room, you need more sophisticated detection.

False positives are the fastest way to frustrate users.