How to composite 3D in flame

[jayfxjay]

Can someone explain how to composite all 3d layers in flame
I mean what do you do with specular passes, beauty, highlights z depth passes… how to comp it
When you go in action with front back matte you always have a thin dark line on the edge, I assume you get best result whith logic ops but WHY
What is the method

Thanks

[kingcoon]

changing the blend mode to add should help with the black lines.

hth tom

[Jeffrey]

Yep I know that, in fact if you want to do it well, you have to invert your alpha then multiply it with back then add your premultiplied front with the result. Then you ge the perfect integration no dark edges. It’s described in COMP 1O1 on this site

But why do we get those dark edges with front back matte in action, that was my first question

And my second, how do i comp and what is the purpose off all thoses 3D passes

CHEERS

[kingcoon]

well if it is rendered plain on black (with no “overprint” – for some reason I never know which one is the pre- or unpremultiplied and too lazy to read right now) you just use add for the beauty layer.

The others are to give you control over the image and composite it really good. So the specular layer gives you the speculars to strengthen them, reflection gives you the reflection. Often simple add or screen (bad control) can be a good start for these layers. z-depth is the depth information so you can add a defocus if necessary – unfortunatly this is a little tricky as FFI doesn´t recognize depthlayers – only as a greyscale image. Shadow passes ar for the shadows on the ground or on other objects – btw a shadow is actually never black! It is the darker variant of the original color. And so on. Sometimes you wil use all the layers and cry for more and sometimes you will only need beauty, specs and shadows. My personal opinion – first of all high quality renderings by the 3D guys and then as many layers as you can convince them to – as said you might not need them all but it´s nice to have them. In the end it has to look great to you (well, and the client has to love it).
tom

[McArdell]

You will find that you have a lot more control comping these elements with logic ops in batch than in action because while action has some transfer modes they are often limited (like some have no opacity control). CG will give you these layers to finesse a comp and most of them are designed to be added, screened, multiplied, etc to the beauty. Logic ops in batch is great for this. Sit with your 3D person and find out how the elements were intended to be used.

There was a technology demo at NAB that showed the layers being generated in Maya and then building a comp automatically in Toxik using information passed from Maya as to how they got used. It just saved the manual part of building the tree and let you get right to tweaking.

We have also spent some time on this type of compositing in courses on fxphd.

Jeff

[Valter Battioli]

you can just divide the passes by the repective matte and apply in action or logic ops, where do you feel more comfortable. Its one step that solve the edges and let you free to do whatever you want (cc, blends, logics ops…….)
About the number of 3d passes, its depends of how much control you want have over the general compositing work.
Hope it help you.
Joao Mineiro

[yogesh Sail]

yeah, click the “divide” button in action or divide on the desktop using the logic op.

one thing I have noticed, and this might be the issue the guy is having, is that if you have your layers in 3d space and the Z buffer on, layers do get a harsh black edge where the matte is gray and not full black or white.

the fix to that is once you get all your layering right, you hit the “analyze” button in the layer stack editor and it figures out what’s in front et al, and then if you shut off the Z buffer everything looks nice.

there could be a better way. I remember when particles gave me the same issue I could just hit “sort” under the surface/geometry menu and that’d resolve it. haven’t found a way to do that w/ regular image layers…

[noo321 ali]

Just a quick note regarding controlling action layer blend modes with no opacity, you can color correct the matte to increase or decrease the opacity, handy workaround

cheers

[peter duncan]

Bonjour jayfxjay,

cela fait un moment que tu as posé ta question, mais peut-être que ce complément d’information te sera quand même utile…

jayfxjay wrote:

When you go in action with front back matte you always have a thin dark line on the edge, I assume you get best result whith logic ops but WHY
What is the method

The main reason for the edges you observe is that Action (or DVE) ALWAYS assumes that the images it manipulate are not premultiplied. When Action does a comp, it will therefore invariably multiply the layer matte with the layer foreground before adding it to the background, which will result in the edge artefacts you observe, if you are working with images which are already premultiplied.

Premultiplied images coming from 3D are very easy to recognize as they will invariably appear with perfectly anti-aliased edges over a black background. Unpremultiplied images usually appear with very wierd jaggy edges which extend beyond the object, while the matte presents smooth edges and transparencies. They can also present a uniform colour around the object to ensure better blending with the background plate.

Although Action expects unpremultiplied images, it does have a blending mode which allows you to work with premultiplied images:
“Add”, also known in the old keyer as “punch back only”. This mode will short-circuit the matte multiplication with the foreground, and add the foreground “as is” over the “punched” background (“punched” means the background plate is multiplied with the inverse matte of the foreground). Of course, if you use it with a foreground layer which does not have black outside of its matte, it will generate an additive effect on the background.

A lot of people don’t like working with “Add”, because the transparency chanel no longer works the expected way (100% transparency results in 100% additive mode – not full transparency). This can be easily overcome with a small trick. While most people use a CCorrect to achieve transparency with “Add”, there is a much more interactive method which does not require the use of an indirect module such as CCorrect within Action (using indirect modules in Action is very memory and CPU intensive).

It uses a very simple expression. The steps are the following:

-go to the chanel editor and copy the transparency chanel of the surface
– expand the diffuse chanels of the surface
-link the transparency to one of the rgb chanels
-edit the expression by adding “100 – ” in front of the transparency variable
this effectively substracts the transparency value from a default diffuse value of 100
-copy this identical expression on all r,g,b diffuse chanels
-make sure that your surface is set to “Ambient” and NOT “Diffuse” in the surface menu

you can now adjust the transparency normally (as you would in “Blend” mode), and interactively, without going into a CCorrect. It also renders very fast (much faster than using CC), even at high resolutions.

As someone else mentionned in this thread, you could also chose to stick to the “Blend” mode in Action and get rid of the black edge by selecting the “Div” option in the layer menu. This will effectively restore (although not perfectly) the edges in the foreground by dividing the premultiplied source with its own matte. You should see the dark edge disappear even though you are still using “Blend”. The end visual result should be very close to the previous result using “Add”.

Note that the layer Divide option is not hardware accelerated and therefore will render slowly compared to the “Add” scenario.

I hope this clarifies things to a certain extent.

Améno

La conscience doit être parménidienne,
l’ignorance et l’inconscience doivent être contingentes…
Mais dans quel état j’erre?

[Michael Schlesinger]

Wow… very clear explanation and brilliant tip, but wouldn’t be easier and save a lot of headaches to just add a “premult” button for each layer in Action. I’m just having to deal with that and takes a lot of time to do that for each layer. That is my wish for next version, I’m alone with this wish? Do people like to have to do all those steps to comp some 3D? For some reason the “divide” button do not work fine if I add a color corrector and adjust gamma or curves before get into the action layer….

[McArdell]

The divide button in action is your premult button – but it is in the wrong place because it assumes that your fill and matte are mathmatically identical.

To see this take your CG fill and matte and make it layer 1 of action with a grey background (do this in batch). Blend mode… no adjustments. Notice black edges around CG in result. Now hit divide in the layer in action… should be ok (if not there may be a CG problem). Now add a gmask on the matte layer (like chop it is half). You will see in new result that this now has a white edge around where you split it. So while Divide may work fine for a simple thing – chances are it will bite you because of where it sits in the order of operations.

Instead leave the divide on the layer off and divide your cg by the matte before action… now you can modify matte or fill without worrying about edging.

Jeff

[Michael Schlesinger]

I’m going to answer to myself…

I think I know what the problem is, I’m using a CC to correct the front layer before comping, and I’m lowering the blacks, so that kind of shrink the border of the object, and that create a black border, even if it is divided after in Action because the Alpha did not had the same treatment . I’m not used to this kind of problems since all the effects on the DS where I use to do all my comps, are RGBA and when I do the same, it apply the same to the alpha, so it always looks good. So I’m guess if I copy and apply the same CC to the alpha channel it will get rid of the black border. Just theory… have to try it. Still can’t get used to have to duplicate the effects when I need to affect the aplha too… 😳

[McArdell]

If you divide the fgd by the matte and then color correct that divided result and leave the matte alone you should be fine.

Jeff

[Michael Schlesinger]

neonmarg wrote:

The divide button in action is your premult button – but it is in the wrong place because it assumes that your fill and matte are mathmatically identical.

To see this take your CG fill and matte and make it layer 1 of action with a grey background (do this in batch). Blend mode… no adjustments. Notice black edges around CG in result. Now hit divide in the layer in action… should be ok (if not there may be a CG problem). Now add a gmask on the matte layer (like chop it is half). You will see in new result that this now has a white edge around where you split it. So while Divide may work fine for a simple thing – chances are it will bite you because of where it sits in the order of operations.

Instead leave the divide on the layer off and divide your cg by the matte before action… now you can modify matte or fill without worrying about edging.

Jeff

Thanks!! have to try it… I’ll let you know if works with what I’m doing. But goof to know about the importance of doing it before getting into action.

[Michael Schlesinger]

neonmarg wrote:

If you divide the fgd by the matte and then color correct that divided result and leave the matte alone you should be fine.

Jeff

Thanks man. So good to have people like you answering so fast!!!

[Author]

Posts