Currently Online 
haxxiy on 15 February 2024
| TallSilhouette said: Generative AI takes its next huge leap forward. Text to video is now actually decent: Probably still a ways off from being used for proper storytelling as opposed to stock footage/b-roll, but the leaps and bounds this tech is making in such a short time is simply staggering. |
The limiting factor is RAM here, you can fit billions of language parameters in an 80GB Hopper for text training but video would be orders of magnitude more complex. Hence all the large language models doing TtV are adopting a mid-term by using their image model and animating based on that. It should be pretty easy to break and make some nice abominations with it :D
