Flame Machine Learning Timewarp, now on Linux and Mac

Hi All,

Here is the link to the Machine Learning Timewarp tool in Flame that @talosh (Andriy Toloshny) put together. It’s open source, linux only (tested on CentOS 7.4 and 7.6) and provided as-is, so use at your own risk.

Unzip the file (1.4GB) and put it in your /opt/autodesk/shared/python directory. When you first launch flame after installing it you will get a prompt that it needs to do some setup tasks. You’ll get a prompt when it is finished. Then right-click, select FlameTimewarpML and you’re good to go!

Please report back the results you get. It will be great to see how this works!




Hello Andy,
the Timewarp AI is a real Gamechanger. There have been many days in my past that I had to paint artifacts from Crappy Timewarps. Sadly its just for Linux right now…are there any plans to implement the tool for mac?
Much Thanks and best Greetings.



Hi Sebastian,

Thanks for reaching out! From what Andriy has explained, this particular model requires CUDA libraries? Which ties it to nvidia. All he did was take the open source library and wrap it in a python script to handle the I/O. If something is released in the public domain that works with Metal on Mac it may be possible, but I don’t know at this time.

Welcome to the forum!


1 Like

Thanks Andy,
maybe I should just think about putting together a Linux Flame instead of a Mac, but that’s another discussion anyway.

It’s nice to finally be here in the forum. It was about time: D

All best,


Welcome @Gummihuhn !

Ran it tonight on a 4K open gate clip at 16% of speed. Lots of hair and a big lens flare over a woman holding a surfboard. Also ran the same clip through motion estimation in flame. While it wasn’t perfect, the ML version beat the ME version hands down. It wasn’t even close. The ML version had only one or two instances of notable tearing while the ME version tore from start to finish.

Impressive to say the least…


On my side, I can’t run it.
It gives me back an error.

What version of CentOS are you running?

This may be unrelated, but what version of the cuda drivers do you have installed? type nvidia-smi in a terminal. I’m running 10.2

Not sure how to check exactly what version of Centos it is but it’s a 7 something.
And my CUDA driver is 10.0

I’m on flame 2020.3

cat /etc/centos-release

I’m on 7.6.1810 and CUDA 10.2 (Flame 2021 on a Z820)

1 Like

Ok will check tomorrow.

So I’m on centos 7.4.1708

if you’re on mac, it is coming nowish, I’ve seen this error on mac recently. It could probably mean that you have a newer Nvidia driver as well, neet to investigate more, could you run nvidia-smi on your machine?

1 Like

I did run Nvidia-smi already. My driver is 10.0.
centos 7.4.1708, and Flame 2020.3

Is it saying that cuda version is 10.0? This build is done on 10.2

Ok, that would explain everything. I’ll check with IT if they’re ok to update Cuda drivers for me.
I’ll let you know if it fixes my problem

Are you connected to the internet on this machine? You may try to install Miniconda and packages manually and check if it is going to work

I don’t really have the autorisations to do that on my own. Everything I want to do, I need to ask IT. But I will!
First update Cuda Drivers

This wouldn’t need anything more on the system, just user-space thing, that’s already been done by the script. You won’t need root password for that

Ok, then, would you tell me how to achieve that?