Why do NLEs need to render a clip again after one cut?

[Gwendolini]

Hi,

I am using FCP X nowadays, but have seen the same behaviour in Media Composer and Premiere Pro.

I have recently edited a small theatre performance, where I have done to more complex edits for two pieces inside it and then have done the entire performance in one camera angle with the two pieces inside.

Since the camera used was quite noisy, I had to apply the Neat Noise Removal plugin (wonderful piece of software) and that took quite long to render the 60 minute long sequence.

But as I wanted to change the last 60 seconds and switch to a new camera, I made a cut at 59 minutes and changed the angle and FCP X now has to render the whole first 59 minutes of noise removed footage again.

WHY? This is crazy. I sent feedback to Apple, but they will never answer my confusion.

Do any of you have a technical answer why an NLE has to render an already rendered clip again just because 1% or 99% of it gets cut away?

 

[Krevnik]

I can't answer definitively, but making cuts to a clip changes how keyframes/etc play out in the rendered version. Shortening the tail end of a clip shouldn't be a problem, but splitting, removing the front, and other changes can make using the previous render difficult while showing you what it will actually look like. Playback has to start at a key frame, and splitting a clip in the middle of a sequence between key frames should require the second part of the clip to re-render.

I suspect the apps mark the clips as needing a re-render based on certain changes, and aren't checking specifically for the small handful of cases where it is safe to not re-render. I also suspect that people using single long takes with expensive effects applied are rare enough that they considered the trade off worth it.

 

[Gwendolini]

But what if every frame is a keyframe? I always use ProRes optimised footage.

 

[LethalWolfe]

I can't say for sure, but if you rendered the entire timeline at once then FCPX probably created one big render file. So when you made changes to your timeline FCP had to throw away that one big render file and make another big render file.

If, for example, you rendered your timeline in 15min long chunks then your change at the end probably would've only required re-rendering that last 15min chunk.

 

[ColdCase]

Is the 60 minutes one large clip? If its broken into smaller pieces (cut with blade tool) then it should only need to re-render the smaller piece affected. You see this a lot with certain filters that inspect many adjacent frames to optimize the current frame. So all frames in the clip can become dependent on all the others in the clip. Change a few frames, and the entire clip needs to be reanalyzed. Dunno if thats how your noise filter works, however.

 

[joema2]

....Do any of you have a technical answer why an NLE has to render an already rendered clip again just because 1% or 99% of it gets cut away?

The NLE does not always do this -- it's dependent on the requirements and characteristics of the effect. E.g, you can apply color correction to a clip, cut it and it won't require re-rendering. Color correction is an intra-frame effect -- it is applied uniquely to each frame without reference to the surrounding frames.

However other effects such as video noise reduction, stabilization, etc. are inter-frame effects. Analysis of the entire clip is required before applying the effect. If you cut off the end, you have changed the conditions for that analysis.

This is easier to visualize for stabilization than for noise reduction. Suppose toward the clip end there's a big shake which it incorporates into the overall stabilization plan. It formulated the applied stabilization effect for all frames taking into account that shake at the end. If you cut that, off the previously-calculated effect is no longer valid and it must be re-rendered.

With video NR it is calculated and applied using temporal or inter-frame considerations. If you cut a piece off you have similarly invalidated those calculations which must be re-done.

I'm not saying there is no possible future method to optimize this but that's how it's commonly done -- not just for Neat Video but other similar effects.

In general the best practice for time-consuming effects like stabilization, video NR, flicker reduction, etc, is deferring those to the absolute last step after all other editing is finished. Don't apply them early in the edit sequence. Even in Hollywood movies it's done this way. When you see alternate takes or early versions of a film, they are typically uncorrected for sound, color, etc.

In case you haven't done it, Neat Video can be optimized for your particular hardware using this menu option:

Within Neat Video dialog: Tools>Preferences>Performance>Optimize Settings

That will show you the relative performance of all combinations of CPU cores and GPU and let you pick which one has best performance. However nothing will make it run really fast, as it's doing a tremendous number of computations across many frames.