Make Pixels Dance: High-Dynamic Video Generation | Hacker News Reader

Make Pixels Dance: High-Dynamic Video Generation

makepixelsdance.github.io

96 pointstrueduke16 comments

So my company says "This site is blocked due to a security threat that was discovered by the Cisco Umbrella security researchers."

Anyone have any idea why?

ByteDance is the company behind TikTok which might be blocked by your company, try loading a TikTok URL.

when I got into the office I tried it again and I got more info -

This site was blocked due to the following categories: Newly Seen Domains

pretty weird.

opening up the description of the site says Host: malware.opendns.com

malware.opendns.com is also blocked, which in my experience here they will block for silly reasons. I wouldn't be surprised if they blocked anything not whitelisted that had the word malware in the url.

Do any github.io pages load for you? Might be that they got reports of a github.io page being used to link to malware, and decided to block the entire domain.

https://artsy.github.io/ works fine.

ByteDance Research working hard to generate all future content on TikTok via AI

They have the entire world generating content on their platform for free, why the heck would they ever give that up?!

They can control the entire narrative.

If the viral spread of the letter was the weaponization of tiktok, now they don't have to convince people to even do a "challenge" just make the videos. If we can't tell if it was made by a real person or not, isn't it all the same?

We need digital fingerprinting of people generating their content to validate they are not computer generated. This is a good use-case for blockchain.

The content is already heavily moderated and controlled, for example the content you receive in the Western world and in China is drastically different [0].

[0] https://www.deseret.com/2022/11/24/23467181/difference-betwe...

[Citation Needed]

They won't give out the model due to "Security Risks"

On their example page you can sometimes see what data it got to train on. https://imgur.com/a/CuXNATp for example is without a doubt "Ragnaros" from the game "World of Warcraft".

Probably due to not having much different sources for "A fiery humanoid monster with flames burning".

I guess Pixar will be safe for a while. While text-to-image models work pretty well, text-to-video is pretty hard to look at, judging from the example video.

when dalle2 was announced in 2021 the text to image models were pretty bad, and now we have models like midjourney that are almost indistinguishable from real images. If I've to bet I would expect the same progress from text to video models

The sample video with a lovely story for children looks good. This has the potential to bring down costs in content production

Lovely story.

Reply on news.ycombinator.com