Cineversity

Hi,

So I am wondering is it better to use straight alphas ( 32bit LWF ) or create a colour ID pass in C4d for colour keying in After Effects. Colour ID is definitely quicker but what are the pitfalls of using a colour ID as opposed to Straight alpha?

Thanks

Hi Kiwi,

I hesitate normally to give short and general answers about compositing Multi-Pass. It can fill books to cover the majority of pipeline set ups. But all in all they are not so complex.
However, there are too many variables and you need to know what you are after. I’m certain as well, that I haven’t tried all options yet, even I try to set up each workflow differently to gain data since years or let’s say way over a decade (Shake, NUKE, Fusion…etc). Especially now with REC 2020 or more in my area DCI-P3, anyway, UHD-HDR and perhaps even more will change how we do things. Hence why I’m careful in answering, since we are in a constant change.

Two things are pretty simple and clear, Straight-Alpha is not something that can be seen purely as masks only creation, even sometimes it is applied just like that. It is often not possible to get really straight RGB values from a scene, hence the idea to render things separately for critical elements.

The Straight “concept” means that each semi transparent pixel is rendered as if the “geometry would cover the whole pixel. In this way any color correction or other adjustments will affect the whole RGB value, not only a partial value. If you look for hi-end work, this is the way to go. There is no replacement with simple alpha options. After the corrections is done, the values can be “merged” or/and premultiplied if needed. This might already answer the question.

Any way of color coding has no connection to the RGB [image] creation of the PIxel value itself. Which is normally the idea of Straight, to have all pixel values for that object, if possible, things are often mixed of course, based on many parameters.

The following is just my opinion, not really proven, as I don’t work that way: [subjective]
I wouldn’t really key it, if you like to have more “masks” in one pass, have each channel as one object only: R, G and B. Anything above that, wouldn’t be my way. Take the information as is; This is the fastest and cleanest way. Any mixture of color might be nice, but pulling a key for many objects might slow down the pipeline more than your win. [/subjective]

The concept of Open EXR Multi-layer files in NUKE is certainly very advanced,  in Fusion this concept is not as far developed for example; Not even to think about other apps who barely reach the 32bit/float per channel pipeline, speaking of all tools. So if you need to key things, which I wouldn’t if possible to avoid, is there a 32bit/channel float key at all available in the target app?

The question with color coded passes would be always, is it easy to set up, and in case of adjustments, how much work is it to replace and control the replacement. thinking of close to deadline scenarios….

I’m certain that there are many other ideas about available. Since Open EXR compresses very well, all parts in a matte/alpha that is black will certainly not take a lot of date (run-length compression), so why merge it. On the other hand, why calling up a larger file, just for one channel. Which is as well in contrast with the multi-layer idea at all.

All in all, I think it is at least worth a studio based discussion. Then stick with it for a show or at all. The certainty and low amount of communications based on standards is perhaps better than to optimize the files but waste a lot of time.

Since we have in C4D the Take System, I do not really see a need to place too much in one file, but of course I see the multi-layer file as something valuable as well.

So, the answer is - it depends, even I think it is an old concept, but that is just me. ;o)

In all cases, the color coded matte combination file needs urgently be protected to get a accidental gamma treatment (it shouldn’t , but since it might be confused with an image file…), which might be easier to sort out with one channel files.

All the best

Thanks Sassi

So rendering out straight alphas is going to be faster, or as fast in the long run if you have the performance hit of keying out all your ID passes in your app of choice. Makes sense.

Kiwi, you’re welcome.

If it is faster, I don’t know, as I wrote it depends on many parameters. If you have many thousand objects and one needs quickly access to a single one, there would be no doubt that it is faster (see image, with 3.000+ objects), if it is the better quality, well that needs to be explored. With many objects in the scene, a manual quick-mask (trash-mask in comp app) might be needed, as edge values might be similar to other object colors then. Where the sweet-spot is, is not clear to me, as colors are given randomly.

The best compositing results are given if the objects that will be covered by the mask will be not “contaminated” by other objects. Think of the the border between two objects, one is masked and one not. The shared edge is antialiased. Which means that the edge even masked will give no full pixel values. In extreme cases, like with very a very dark and a very bright object creating this seam, the antialiases will mix the pixel-values. The mask for one object, in any way, will not have the original values (as if the object would have been alone).
This is crucial to understand. If the object in question is now blurred, color corrected or in any way altered, these mixed values might show up. Especially with HDR values, as AA it not really set up for values above 1.0 [or 100%]

I have tried to explain it here
http://www.cineversity.com/forums/viewthread/1948/#7423
with a formula example.

A typical way of compositing elements is given with a “Light-Wrap”. An outline mask is created and then the image background [eventually the pre-comped] blurred and only then a blurred outline of this image via soft outline mask is added on top of the image. It has a little bit to do with the “fresnel” of any object, and the light that might affect the foreground from the background in the area of the rim. (Discussed in my JET series). Most normal alpha problems might be suppressed to a certain degree in this way. But it is not suggested to get sloppy with the masks, just because. In strong Motion blur or DOF blur areas it is kind of a judgement call how to apply it, if at all of course in those areas.

The Straight vs normal alpha is discussed in my Multi-Pass series. All in all a question of quality, how fas one likes to push it.

So, my few examples might clear a little bit that each process and target shortens or extend the time needed.

My best wishes.

P.S.: 

In any way you would create a color object matte, keep in mind, that the colors of the objects determine the Anti Aliasing values, if set to an AA Threshold. As mentioned above, the AA mix of object colors might lead to colors given in other places of the color matte.

This threshold is based on the difference that the color of the object has, I’m not aware that it will take any other values than that. How to meet then an equivalent to the image based passes is certainly not simple.

I’m also not aware how much those techniques allow for transparency and motion blur mix (fg vs bg) for that matter. Again, as I mentioned, it requires a longer discussion, to find the best way for your needs. Any general idea about—might work only in some cases.

Also, the colors should be consistent and not randomly distributed, otherwise a re-render will require to “key” all of these. Especially if you go beyond the three [RGB] color set up, while thinking about motion blur and depth of field blur, with the Physical Render for example. Sometimes the present situation provides already the decision.