[Spencer Tweed]

Owch, looks like he no longer has it up for download! I don’t know if it is windows-only or not but here’s the copy I have (i’m on windows).

12160_spillsuppression.aex.zip

Otherwise I haven’t tried the keying tools in CC but I think there’s a newer spill suppression plug-in that might be just as good. The Cinegobs one allowed you to generate a matte from the spill which was WAY softer than your average key. You can do it with keylight I suppose, just color-sample your green screen and leave the settings at default in terms of the matte, then set your “view” mode to “Screen Matte” and invert it. I suggest you denoise your footage and precomp it for anything relating to matting.

Gimme a still frame and I’ll throw a project file together. It’s taking me longer to explain this than it probably would to just set something up…

– Spencer

[Spencer Tweed]

Hey Stephen,

Believe it or not, you can use the techniques in that tutorial to get your blurry key ???? It’s actually the same problem – you’re just trying to get a “soft key”. Don’t take the tutorial too literally, just see how you can apply the techniques to your footage.

That said I will give you one other piece of advice. A lot of times it isn’t the matte generation that is the issue, it is actually how you treat the edges. PARTICULARLY with blurry edges, you’ll actually need to mix in the footage into your foreground BEFORE you matte it on top of the background. It’s takes a lot of precomps, which is why I use Nuke… Anyway check out CineGobs “Spill Suppression” which is a fantastic plugin I could not key without. First you’re going to want to use that plug-in to despill the foreground. Then separately you can actually create a SUPER soft matte with that plugin (I wouldn’t even call this a key because it is quite soft and dirty), which you use to overlay your background plate on TOP of your foreground greenscreen footage. Precomp the whole thing and THEN matte it out by your original matte, and put this precomp over the background plate. Lastly play around with how much you blur the background plate before you overlay it on the greenscreen footage.

Whew, hopefully that makes sense… Let me know if you need a step-by-step.

– Spencer

[Spencer Tweed]

Absolutely and great question! Check out this awesome video from Andrew Kramer covering this exact issue: https://www.videocopilot.net/tutorials/advanced_soft_keying/

And lastly it is worth mentioning… I know this is an AE forum but I haven’t actually used After Effects to key anything since the day I picked up Nuke. I couldn’t suggest it more. And if you do go that route, Mr. Lyons is the master of keying: https://compositingmentor.com/. That’s some pretty damn complex stuff, but if you really wrap your head around it you’ll be doing ILM-level keys!

– Spencer

[Spencer Tweed]

That has nothing to do with what he is asking…

[Spencer Tweed]

Thanks man! I struggled with this same issue for years, and he had a valid question.

For your generous acknowledgement I’ve got some code to post up here for free ???? It’s been a while since I posted any extendscript anyways. Want my “background instancer” which launches background AE processes so you can instance render while still working? I wrote one ages ago and never released it. It’s windows only (I don’t own Mac) but if you are on PC it’s a life saver, and got me by for years until I got a proper render manager.

I’m mid a major project right now so it might take me a day or two to package it up and post it. If you want it and don’t hear back just send me a nudge.

– Spencer

[Spencer Tweed]

I don’t remember exactly what I ran into, but I got confused because of my recursive loops and whatnot. I’m sure if I took the time I could figure out the profiler, which I’ll do when I have a moment.

I got an interesting reply on Stackoverflow which doesn’t use splice: Link

Only problem is I tried running it in AE and I got an error on the array.replace() method – and in doing some research looks like Extendscript doesn’t support higher-level array methods… Reference. However per that post you can add them with Polyfill, but I haven’t had a chance to try it out yet.

Anyway, basically just some more testing to do and no time to do it!

I’ve uploaded my script if you want to have a go, or just use it for free. I use it non-stop throughout the day, I don’t even import files the old way anymore unless they’re single images or video clips (hence why I’m trying to optimize). I continue to develop it every week or two, but in it’s current stage it is totally stable so far as I know. That said i’m still on CS6 for the majority of my professional work, so you might run into a thing or two in CC…

10494_stimportpanelv1.2.jsx.zip

– Spencer

[Spencer Tweed]

Lens info absolutely. But truthfully it all depends how far you’re planning on taking this shot. If you want to go all out you would do the following:

1) focal length of that shot. Avoid zooms but if a must, at least get the starting and ending focal length.
2) film back size. Or at least note exactly what camera you’re using so you can look it up later.
3) lens grid for undistorting your footage. This goes beyond the scope of After Effects but if you’re going into PFMatchit, Syntheyes or anything heavy-lifting like that it can seriously improve your track. If you think about it most lenses distort at least a few pixels (if you’re lucky). Well imagine all of your tracks drifting a few pixels! It can make a difference. And trying to undistort footage without a lens grid is, in my humble experience, often worse than just leaving it (particularly on greenscreen where you have very little data to work with).
4) distance from camera lens’s nodal point to a key object.
5) height from ground to camera lens’s nodal point.
6) survey data – such as object sizes and distances.

The last two are only really needed if you are going to be re-creating your set in 3D and trying to match 3D objects in. it doesn’t actually improve your track very much unless you’re going ILM style with LIDAR scans and stuff. Which I seriously doubt… but would be pretty damn fun to play around with. Anyway the AE cameratracker doesn’t have any use for real-world survey data unfortunately, so you can probably skip it.

Good luck!

– Spencer

[Spencer Tweed]

Nice! Last time I did this I just glue-stick’d a printout of the marker onto a piece of cardboard. Your method seems much more professional 😉 Just make sure your paper is not reflective or you’ll have problems. Trackers typically hate luminescence changes, so the more matte the better.

– Spencer

[Spencer Tweed]

Thanks for the tip, there are definitely other ways of doing that step so I’ll check it out.

On the topic of profiling, what do you use? I know the extendscript toolkit has a built in profiler but I found it quite confusing, and I use sublime (can’t stand the toolkit). It probably doesn’t help that this is all wrapped within a recursive function… I tried out $.hirestimer per the manual but I found some funky stuff – for example if you run the same script back-to-back you will get a much quicker timestamp on the second iteration, which I assume is due to the javascript buffer but $.gc didn’t help.

– Spencer

[Spencer Tweed]

Hey man, long story short AE is surprisingly slow to render sometimes. However there is a way to basically 20x your render times, but it gets a bit complicated. Sounds like magic, but it isn’t. Here’s my really long answer, documented here in case anyone else can also benefit from it. First the theory, then the how-to:

History: AE’s renderer was built back in the 90’s I believe, and hasn’t kept up with modern multi-core computing. Adobe knows this well and has been trying to fix it, but their first attempt (multi-processing) failed due to over-complexity, some finicky coding, but mainly just users not knowing how it works (admittedly it wasn’t very clear and certainly wasn’t easy to work with). That said they ax’d multi-processing in CC and decided to just do it right, and hence have been mid re-coding AE from the ground up for the last few years.

Theory: As previously mentioned, the renderer doesn’t work with modern CPU’s very well at all. The evidence of this will be clear for yourself shortly. What happens is that AE bottle-necks somewhere along the line while rendering and spends probably about 90% of your render time waiting for things to happen instead of actually computing – if you don’t believe me then turn on your resource manager while you render and watch the spike every few seconds/minutes and the 2% idle in the meantime. So the way we can fix this and leverage all of our expensive 2016 computing power is by rendering multiple INSTANCES of After Effects at the same time – on the theory that while one instance of AE is sitting there waiting for that bottle-neck another one can jump in and start computing. It works shockingly well and is loosely documented here: https://cgi.tutsplus.com/tutorials/significantly-speed-up-your-renders-from-after-effects–ae-26822

The technique: Personally I use Thinkbox’s Deadline to render on 3 machines, and I’ve never had a render take more than 40 minutes (that was for a 2600 frame shot btw. 90% of my renders take between 2-10 minutes). I’m guessing you don’t have an easy way of setting up Deadline though it is free for 2 nodes, so I’ll just skip that and move to the at-home free method. Luckily you’re on a Mac so it is a bit easier for you.

1) set up your render queue properly: make sure you select “multi-machine” for your Render Settings and any image sequence format for Output Module (I use .png). This whole technique relies on rendering image sequences, if you need a video file convert it later either back in AE if you have to or using whatever you want (I use FFMPEG). Make sure to save your project file, but do not render yet.

2) download this application and install it: https://www.monologue.gr/tutorials/monologue-ae-render-manager/

3) Now, if I remember correctly, just drag-and-drop your project file onto the AE Render Manager icon to launch an instance. I don’t have a Mac and I’m too lazy to read the documentation, so I’m not 100% on that one – but I’m sure you can figure it out. Alternatively as the above video suggests you can pay for BG Renderer Pro and do it that way, but I don’t have a lot of money so I’ve never done it.

4) Launch as many instances as you can safely run. I tend to make really heavy comps and I’m able to do 5 instances while I work on my machine without any noticeable difference. On my render machines I can do 10 no problem. That means I am running 5x-10x render times PER MACHINE! And yes the math does check out! The above video suggest you run one per each CPU core, however I can tell you that will just choke your computer out and leave it gasping for air if you’re doing anything more complicated then a few text layers or color correction – the reason being you will run out of RAM and the over-lap of the CPUs will slow things down. Try 5 and add a few more if you don’t notice any slows, it is a balance of hardware and how heavy your composition requirements are so there really isn’t a fixed number here.

5) You’ll see the frames spitting into your render folder. Once you’re done go ahead and convert them to a video file or whatever.

* lastly a word of warning. I include this not necessarily for you because I don’t think you’ll run into this problem, but for anyone else who cares to read this and needs the data. I’ve been using this technique for a few years now and I’m noticing one thing – effects that render on the GPU will CRASH if you run too many instances at a time. This instances technique was designed to more efficiently leverage a multi-core CPU, but GPU’s are a whole different thing and do not play by the same rules when it comes to CUDA. Software running on the GPU DOES actually use all of the cores, so if you render too many instances they will choke eachother out pretty quickly. You’ll also run out of vram (RAM in the GPU) very quickly. So to fix this I pre-render out things like Element 3D and Plexus. In the coming years as AE continues to be re-written and more and more ported over to the GPU we may reach a point where it isn’t viable to render more than a couple of instances… But in the mean-time it is simply magical and theoretically when that happens everything will be so fast it won’t matter anyways 🙂

Whew, viola! I’m happy to answer any questions.

– Spencer