Flame Machine Learning Timewarp, now on Linux and Mac

Hi guys, those who have access to flame 2024 on linux - could you try to download and check if it works? Please unpack and put the whole folder to your python scripts.

8 Likes

It looks like this for me:

1 Like

This looks to me like it is actually working but unable to pull the image from clip. Is it something that is happening on other clips as well? It is not needed for a clip to have a timewarp effect, and the best is to have it rendered if there are other effects

Thanks Andriy!

Just to confirm, is this the menu I should see when I right click a clip?

This is what I’m getting when I hit Timewarp ML → Timewarp on a clip with no effects:

By the way, if I hit “Close” on in the lower left corner, everything is fine. But if I hit the “X” in the upper right, I can’t access the contextual menu or the Main Flame Menu. Let me know if you want me to try anything else.

Thank you for your help!
It looks like it does work on 2024.
Would you mind to try changing the speed value in the field that says 100.00 into something and try rendering it?

I’ve seen this behaviour when closing window as well, will try to have a look into that. “Close” button is mapped to a function that should wrap up everything nicely and copy the resulting clip next to the one you’ve run it

Installed without issues (had to fetch two python modules, but that would expected).

Loads fine. Ran ok on first clip at 50% until I stopped it. Ran it on a second clip and crashed Flame:

File “/opt/Autodesk/shared/python/flameTimewarpML.v005dev004.linux.flame2024/flameTimewarpML.py”, line 2553 in _update_interface_image

Application: flame 2024 x86_64
Pid: 2495140
Error: abnormal termination, signal = 6
SIGABRT - Abort

Restarted Flame and ran same clip second time this time at 90%. Ran for 214 (of 500) frames and the crashed again. Clips where Arri 35 4K footage.

Also noticed, that when I add speeds of > 100 (in my example 400), it runs through and says it rendered, but didn’t seem to do anything. Top right ‘To Frame’ stays blank during render, no clip is created.

Thank you!

The thing with the packages is a bit strange ‘cause the folder has the packages in flameTimewarpML.v005dev004.linux.flame2024/packages/.lib
And it should have been able to find it and load from there…

It had a pop-up that I needed to load numpy and torch, which I had to run the commands for. MLTW was never installed previously on this system.

Sure thing. Here it is, working perfectly from what I can tell:

Hello :slight_smile:
Where is the last version of TimeWarpML for Mac please ?

Hi Shiver, let me check if v0.5.0-dev-004 are fine on Mac I can not give you an ETA but it was ok on Linux so far so it should not take too long.

2 Likes

Hi, I got my new M2 Studio today. Can someone help me find the best version to install ML timewarp please. I’m a bit confused with all the versions on GitHub

Grab the latest and greatest. v.0.5.0 dev 004

Install Alpha with “potentially lots of issues” ??

You didn’t ask for the most stable version. :slight_smile: I mean, if you’re gonna Beta test Flame, you might as well look at the alpha of Timewarp ML.

1 Like

you know what my answer is to that :slight_smile:

Hi Johnag,

I might not have an access to a flame on Mac - need to check if the latest 2024.2 beta still works after the end of a beta cycle.

In general the problem with Mac was there was no hardware acceleration implemented in PyTorch - the backend used by TimewarpML and the calculations was very slow performed on CPU.
This acceleration has been implemented relatively recently and it seem to perform relatively well as long as you have enough VRAM, but I was seeing some weird thing happening when it is not the case, and there was no any indication that its running out of memory, the result was just corrupt.

So, if you really want the old thing running on M2 you might need to pick version 0.4.3 from the releases page and then you’d need to manually tweak miniconda installation in order for it to run on M2. That would allow you to run it old-style on CPU with the same results and its just a matter of render time.
If you have time I might be able to help you with setting it up and then it might go into a new release of v0.4 for M2

With the newer version - there was many changes first of all to the way it interacts with Flame, and then it might try to use GPU acceleration on Mac.
It is in development and I can not give you an ETA on when it might be really stable to use but I’m working ot it slowly and its getting better and better.

Actually both versions can co-exist on the same machine.

Thanks for the info. I’ll give .5 a try soon. When I complete this current job (by the end of next week)

@talosh if you ever need machines for development, I’ve got a few on Teradici or Parsec you can use.

1 Like

Let us know how you go @johnag make some notes for us. You have a fairly beefy machine right?

1 Like