Blackmagic Forum

I have a 23.98 UHD comp with 2 DNxHR HQX sources. This is 2-up interview where I am using the actress on the left from one source (take) and the actor on the right from the second source. There is a polygon mask on the foreground node with minimal animation dividing the screen.

I have used Mocha's remove on the background node to remove her hand where it passes in front of the actor on the right.

• The result is good but it takes about 45 seconds to render each frame!
• I rendered the effect in Mocha but when I go back into Fusion, it still takes 45 seconds to render each frame.

I am not greatly experienced with either program so I may have missed some settings that would help.

• How should I be caching?
• Is this a case where I should be disabling Open CL? My GPUs are 2 x 1080ti.
• Anything that could speed up the render times?

My system:

Fusion Studio 9.0.2
Mocha Pro OFX plugin 5.6

Win 10 Pro 1709 build 16299.547
ASUS X99 Deluxe II
Xeon E5-2696 v4 2.2 GHz
64GB RAM
2 x 1080Ti

Thank you for any help and let me know if I can provide more information that would help.

I'm not entirely sure how the OFX plug-in operates, but I'd guess that you need to bring in a new Loader with the rendered output from Mocha and replace the connection from the Mocha node. If that doesn't work, then prerender the output of the Mocha node with a regular Fusion Saver.

I would prerender it.

Thank you both.

I'll try prerendering and see how that goes.

My experience is the same. I often times export DPX from Fusion to Vegas as Mocha transfers in Vegas are fast.
The other thing you can try doing is exporting from mocha fusion comp files. or simply copy paste Fusion shape data as shown here

Heads up: Depending on your configuration, shapes in Mocha may be temporally offset in Fusion. Double check in the Timeline View that the first keyframe lines up with the first frame of your shot.

Thanks again.

I tried a prerender to do the remove, without the foreground node and polygon and merge. This is pretty much the same operation as in Mary Poplin's video (3rd one Rajiv posted). Unfortunately, there is no increase in speed whatsoever.

In a way, this doesn't surprise me, because the Fusion polygon mask and merge are both trivial.

Unfortunately, though, it looks like this shot is still going to take 45 sec/frame!

Sounds strangely slow if it´s basically one image over another with simple mask.

Mocha Remove is a computationally expensive proposition, but I didn’t expect it to be that slow.

It’s also puzzling that it is so slow even after being rendered in Mocha. I would expect that Fusion would be able to use the work done in the OFX. Mocha chugs through it at a few FPS.


Sent from my iPhone using Tapatalk

If you've done a prerender, why is the Mocha node still connected to anything? The idea is to get it out of the picture so that you don't have to wait on it.

Bryan, maybe I am not understanding.

The prerender comp is just a loader, the Mocha OFX doing a remove, and a saver. But the render of this comp is still taking 45 seconds per frame.

Is that not what you meant?



Sent from my iPhone using Tapatalk

Depending of settings and resolution, Mocha remove is slow and it can be very very slow.

With prerender you are able to bypass that mocha node once it has rendered and saved so make sure to turn off that node after prerender and work on different branch.

Simply read prerendered clip with loader and work with it without no mocha node involved.

Thanks for helping, guys, I appreciate it.

I may not understand what you mean by prerendering.

Bryan, on your blog I find this:

Prerendering is commonly done to "lock in" a complex section of the composite that you don't expect to change. Make a Saver at the node you want to prerender that points to the folder where you keep the elements specific to your shot, disable any other Savers in the comp, and render, preferably to OpenEXR. Bring the new sequence back in with a Loader and plug it in to any inputs previously fed by the prerendered node. Disable the prerender Saver and re-enable the other Savers. Don't delete any of the nodes you've just prerendered—you may yet need to go back in to change something. So long as they're not wired into the rest of the flow, though, they will no longer slow your comp.

This is what I've tried to do here.

Is this what is meant, or is there another way to prerender?

Without knowing what your flow looked like before, I couldn't say. But my point is that if Mocha is what is taking 45 seconds / frame, then the output of that is what you want to prerender.

If the Mocha node's at the very end of your flow, then there's not going to be any help from prerendering anything.

But, if Mocha's already disk cached the clean image internally, as you suggested it had, you should try to find that sequence and import it to Fusion, bypassing the Mocha node entirely. Chances are, that sequence is in a folder called 'results' next to the source footage.

As I said, I've never used the Mocha OFX plug-in. I only use the standalone version. So I'm sort of advising with one eye closed.

Thanks Bryan,

See attachment for the original flow.

It doesn't appear that the OFX plugin creates the results folder, and from what I read on wesuckless, "It's all internally saved in the .comp file."

There is a Export Rendered Clip command in the plugin's file menu, and I tried that, but the sequence it created is not the rendered clip, it appears to be a kind of mask the OFX host is supposed to interpret.

I am going to contact the support staff at Boris/Mocha to see if they have a solution.

Thanks to everyone who's been helping!

Make sure you have turned ON Render module remove from Mocha node. Otherwise Mocha remove won't render that.

Hey there,

Mark, I've read the whole thread for I'm into Mocha too, especially the remove module...

My 2 cents are that one cannot rely on any performance when an ofx is involved anyway, to make it short, so the best way may be to avoid rendering through/by the ofx. To manage it, copy paste track data is indeed a great way because you have the job done in the ofx UI BUT not the render itself. As Bryan added, beware of the potential drift if the start are not correctly made.
You may prefer to export a comp instead (which could be open aside), why not.

But all this put apart, what I wanted to point to is the way you achieve the remove, which is the very first step to check, because even in Mocha itself (standalone), you can make it 5 to 10 times faster if you don't render the whole clip.
See : (very) shortly, just roto the the hand when it passes and render the frame only and check the result > make a cleanplate (bottom left, "create"); then go to the next part and do the same, again and again in the clip (render just one rotoed frame and create a cleanplate).

In the end, tick "use cleanplates only" at the bottom end and then only, render.
Maybe you already did it this way, if not be aware that even in Mocha alone, it speeds up the process a lot.

Thanks Juha and Sam.

Sam, thanks, that is an interesting approach. But I think in my case, it would take me longer to roto her hand in each frame than to let Mocha calculate the remove.

She is gesturing a lot and her hand is flexing, I would have to roto every finger segment separately!

I am using a looser mask, more like a mitten, which only needs maybe 15-20 keyframes for the whole shot.

I do like using cleanplates when I can and I appreciate the tips.

It's entirely possible that this is just how long it takes for this particular remove.

I'm in touch with borisfx support and am sending them my clip and comp for their analysis and opinion. I'll report back.

Ok, as you wish, but I'm still thinking that a couple precision roto save into reference cleanplates would save you a lot of time instead. It can be a looser mask too actually, it's all about computing important frames only...

I see. I think I misunderstood you, I thought you were suggesting a full roto of the hand.

I will try making clean plates of a few important frames and see how it goes. Thanks for clarifying!


Sent from my iPhone using Tapatalk

I must have been unclear, my bad

Because yes, I wrote roto, but actually a larger garbage matte including more than what you want to remove + well positioned blue rectangle shape within it if you track could be enough to make it. If more is needed, than you should maybe open the cleanplate and modify it in PS, but I don't think that should be necessary in your example. But above all, you don't need to track here in this WF

The main idea is to only focus on portions which really change from all the similar frames before > put your mask > render this frame only > make one cleanplate of it > go further to find where the context changes

Then, you now have one cleanplate by group of similar frames and this is what is going to speed up the final render when you click "cleanplates only".