The_Sloth_Sleeps
Arcane
- Joined
- Oct 26, 2016
- Messages
- 2,440
If someone proves something can be done then the genie is out of the bottle. Over time others inevitably also find a way to do the same.Well, you probably won't be able to tell how it's different even after it's out, since "Open"AI doesn't usually release such info anymore.What I am asking is, what is the underlying tech? How is it different.
My bet is that it's a Mixture of Experts video model. So, you have the MoE language model tied into multiple video models that effectively work as one to create a final product.
MoE is how all of OpenAI's models are suspected to work. You train multiple small experts within a model's framework. An expert on math, an expert on logic, whatever. They all work together to create a final output.
Anyone? Not when it's a closed-source SAAS.I am actually a little sad because I think it would be a good thing if anyone could generate video like that. I do not think it will ever get there.
The problem is the value of the product and the fact that OpenAI are currently the only ones to control it. What they said about making it available to filmmakers is what's troubling to me. Makes me think it will only be available with an expensive license, and then only with censorship.
But if the tech can be reverse engineered, as Google and others have done with OAI's text models to some degree, there's the chance the cat could get let out of the bag again, and we might be able to get our hands on a similar model.
However its unclear what, if anything, this is actually doing. It does not look remotely capable of replacing film making.