AI General Thread

[the-pi-guy]

It is odd that that hasn't been solved yet. Like yes it is a ton of data but like maybe just make a 32x32 latent video that can be 10 minutes long and then use current tech to upscale it. IDK

I'm surprised there aren't more attempts at chunking. Take the last few frames of a 10 second clip and do image to video. 

I understand it would likely get unstable over time. But it seems fully doable right now and yet I haven't really seen anyone try it. 

[Legend]

I'm surprised there aren't more attempts at chunking. Take the last few frames of a 10 second clip and do image to video.

I understand it would likely get unstable over time. But it seems fully doable right now and yet I haven't really seen anyone try it.

Wait, couldn't current models technically do that already? Actually I think Grok Imagine does. You can extend the clips indefinitely. I went from 6 seconds, to 12, to 18, to 24, and ended at 30. Maybe grok tops out at 30 idk.

[the-pi-guy]

Wait, couldn't current models technically do that already? Actually I think Grok Imagine does. You can extend the clips indefinitely. I went from 6 seconds, to 12, to 18, to 24, and ended at 30. Maybe grok tops out at 30 idk.

Yeah that's what I'm saying. It's totally feasible now to do something like that, yet I've never seen anyone actually do it.