Jump to: Board index » General » Fusion

Best practice: many compositions, or one with many nodes

Learn about 3D compositing, animation, broadcast design and VFX workflows.
  • Author
  • Message
Offline

rsf123

  • Posts: 72
  • Joined: Tue May 21, 2019 3:05 pm
  • Real Name: Roger Smith

Best practice: many compositions, or one with many nodes

PostWed Feb 12, 2020 1:26 am

I wonder what is best practice for achieving complex Fusion effects that have many components.

In my videos, there may be many dozen polygon nodes, but at any given moment in the video, most would be static. In a given moment in the video, only one or two components might be moving.

The options I face are having few compositions having everything inside it, or many compositions each relating to a small number of nodes.

Which is better practice, in terms of CPU, GPU and rendering: to have (i) fewer Fusion Compositions each having many nodes, or (ii) many Fusion Compositions each isolating one moving part?

Or does it not matter, in terms of CPU, GPU and rendering?
Offline

Hendrik Proosa

  • Posts: 793
  • Joined: Wed Aug 22, 2012 6:53 am
  • Location: Estonia

Re: Best practice: many compositions, or one with many nodes

PostWed Feb 12, 2020 8:20 am

You mean in context of Resolve, right? My random guess is that one composition is faster. Having each comp separate has its own overhead and all the node ops must be processed anyway, so I'd keep them all in one nodegraph unless you need to separate them for organizing: lets say some elements are always present and others are clip specific etc.

A fewty polygons is by no means a high load situation, at least not in comp world. In the context of timeline and realtime playback things are different though, what is low load for general comp might be something realtime playback chokes on. Expectations differ in these cases.

Regarding whether something moves or not, it depends. For simple comps I'd argue it makes no difference except when you use time-based effects like motion blur that need access to other frames, because evaluating the curves etc is very low load computationally. Static elements have the opportunity to cache better but I'm not sure it works this way in Resolve at the moment.
I do stuff.
Offline
User avatar

Bryan Ray

  • Posts: 1274
  • Joined: Mon Nov 28, 2016 5:32 am
  • Location: Los Angeles, CA, USA

Re: Best practice: many compositions, or one with many nodes

PostWed Feb 12, 2020 3:21 pm

I'd typically prefer one comp to many. If you're dealing with several, you will eventually have to combine them somehow—either rendering layers or timeline segments. At that point you're going to have a master comp taking inputs from several input comps. Then if you get a note on one of them, you have to address that note in the relevant comp, and render both it and the master, giving you extra opportunities for error. Plus you'll have the issue of ensuring that the pieces register properly, both temporally and spatially, which is not always as simple as it ought to be.

On the other hand, sometimes it makes sense to do things that way, particularly if the work is easy to break up into pieces that multiple people can work on simultaneously. For instance, this scene is all one long take (apologies for the poor quality, but I'm rather surprised I could find it at all):



Since none of the effects in it are continuous, and few of them overlap, we cut it into segments so multiple artists could work on it at once, then we used a master comp to put it all together. It was a huge organizational headache, especially when it came to maintaining continuity in the rig removals, but there's no other way it could have been done inside the deadline. (Major thank you to freelance tracking specialist Zach Miller, who helped us to keep the cameras in registration.)
Bryan Ray
http://www.bryanray.name
http://www.musevfx.com
Offline

rsf123

  • Posts: 72
  • Joined: Tue May 21, 2019 3:05 pm
  • Real Name: Roger Smith

Re: Best practice: many compositions, or one with many nodes

PostFri Feb 14, 2020 2:00 am

I came across this comment in the Davinci Resolve manual.

efficiency.jpg
efficiency.jpg (67.18 KiB) Viewed 119 times


This does not shed light on whether it would be more efficient to separate the Denoising node into a separate clip going from 80-100, or to (as they suggest) trim the node to operate from 80-100.
Offline

Kel Philm

  • Posts: 354
  • Joined: Sat Nov 19, 2016 6:21 am

Re: Best practice: many compositions, or one with many nodes

PostFri Feb 14, 2020 8:26 pm

Would it not remove the need for a Merge node by doing this? That's a saving there. Also means less cache overhead as well.
Offline

Hendrik Proosa

  • Posts: 793
  • Joined: Wed Aug 22, 2012 6:53 am
  • Location: Estonia

Re: Best practice: many compositions, or one with many nodes

PostSat Feb 15, 2020 9:47 am

There is a small logic problem with either the manual example or Fu processing. If defocus doesn't do anything, essentially passing image through, it should ideally be a no-op processing wise (something for devs to do). But if it does have effect, trimming its range doesn't do what is suggested because one also looses the effect, animated or not, for trimmed range.
I do stuff.
Offline
User avatar

Saad Shah

  • Posts: 87
  • Joined: Sun Dec 30, 2018 5:36 pm
  • Real Name: Saad Shah

Re: Best practice: many compositions, or one with many nodes

PostSat Feb 15, 2020 6:16 pm

First, I am not sure if you are working in standalone Fusion or in the DaVinci Resolve Fusion tab. For standalone, I usually prefer a single comp. It takes a bit longer to set the project up that way, but it takes less time and resources to render the project and make changes if the client requires.

For Fusion comps in Resolve, I almost always prefer to break up the project into smaller chunks, and use the comps in the Edit page for fine tuning or editing. This makes it quicker to set up and start the project but usually takes longer to render.

And this is just my personal and completely non-technical opinion :) I trim everything. Even if a node is being bypassed from Frame 0 to 79, and it shouldnt take up any resources to process, I trim it. And thats just because I am running a very modest spec computer. And I dont trust it haha.

If I had to pick, just one solution for most scenarios, I would say I rather prefer to have one comp than multiple.
Youtube Channel: Gargoyles At Work

Return to Fusion

Who is online

Users browsing this forum: No registered users and 7 guests