ML timewarp in timeline

Flame Flame Questions
1

We are experiencing some issues using ML timewarp in the timeline. It’s taking ages for the model to load, doesn’t give a time estimate just says this could take some time and cannot be aborted.

2

That is common to all the ML tools, and has been noted during various test cycles.

The time it takes is based on your timeline resolution I believe. It shouldn’t be more than 20-30s in most cases. But be careful to add it to very high-res timelines if you’re not sure you have enough VRAM. If it seems stuck forever, check the terminal for error log message. Usually memory related message just show up in the terminal, not on screen in my experience.

Definitely could be better, but I think some of this are limitations of the underlying technoloygy.

3

I believe the time it takes is also very dependent on the TW value. 50% will take much less time than 49%, for example.

1 Like
4

You can always pop up the resource manager and check for VRAM, it is the rightmost bar in my system. It will stall if the bar is up and red…

1 Like
5

Yes, important and not emphasized enough point: The way the TL-Timewarp works, is that it does a binary division and finds one in-between frame. So any speed change that follows the binary numbering system is faster than anything else.

50% requires one new frame.
25% requires four new frames (does 50% first, and then halves each result again)
12.5% requires eight new frames.

But 49% requires many steps to get there.

Now, sometimes we have no choice but to use 49%…

6

This becomes even more complex with vari-speed. It can be frustrating to wait, not knowing how long something will take and seeing no status bar. It would be interesting if we could find a way to “burn” the process without having Burn. I had something complicated with multiple clips and I set it up in batch when I woke up in the morning and it was done when I got to work, and happily was correct.

2 Likes
7

Are you using apple workstations?

8

No, linux in the cloud

9

Copy

I know you’re already doing it but for those that aren’t in that habit, break the problem into smaller problems, which in this case is find a way to pre-process the activity and swap the result into the timeline.

You could also request a more performant cloud system.

This is the beauty of cloud workflow - vertical scaling on the hardware is a button push.

10

Personally, I always pre-render stuff like that, outside of the TL.

1 Like