Compositing tips
by Eric Schaechter
Tips for compositing 3D images into live action
[página disponible en español]

The following workflow is my standard for blending 3D CGI into live footage. The realism of a scene depends a lot on whether the viewer feels that the 3D character was actually there at the time of shooting. Although the tips here are meant mainly for 3D on top of live footage, most tips work just as well on keyed persons (or as you English-speaking people call them: "talents"). The most important things to look at before compositing are:

1- Perspective & Tracking
2- Lighting
3- Color Correction
4- Separation of elements by depth
5- Shadows
6- Other CGI (i.e. smoke, clouds, fog)
7- Focus
8- Reflections
9- Grain
 
 
1 - Perspective & Tracking [top]

Your 3D team of artists should give you a pretty good perspective match, but in scenes that have a moving camera that has not been motion-controlled, it is sometimes easier to track in flame. In such cases it is always a good idea to have them send you a shaded playblast (no textures) for you to check out if the movements are going to work for you before having them use all their processor-power in a render that might not work.
 
2- Lighting [top]

It is the 3D-artists' job to give you an appropriately lighted scene. This is one of the most important things for the scene to be credible. Don't hesitate to ask for a re-render, for it is very difficult to change the lighting on a rendered image. Some productions use a white Styrofoam ball on the set and shoot it for the 3D-artists to have a good idea of light direction and color.
 
3- Color Correction [top]

Due to differences in computer and video monitors, as well as the ease of color correction in flame* sometimes the 3D artists will send us scenes that color-fit into the scene but still need small adjustments. The things to look at are the general color of light in the footage versus the color of your 3D. Most of the time you will need just small adjustments on the MASTER BALANCE and on the MASTER LUMINANCE'S GAMMA and GAIN.
 
4- Separation of elements by depth [top]

The reason why so many commercials aren't convincing is that the CGI seems to have been put on top of the footage rather than into it. This happens when you forget to bring fore elements in front of your 3D, or when you don't generate shadows on your 3D (coming from other elements on the scene).
 
5- Shadows [top]

A very important thing about shadows is that they interact with the scene in a very 3D way. We have to study reality around us all the time (that is, if any of us ever sees daylight <g>) to be able to recreate it in our work. Here are some examples:

Remember a shadow on the steps of a stairway? It breaks with every single step, yet the outline could be traced with a pen without lifting it. If our 3D friends can easily generate the shadow, we just thank them, composite and go to sleep a bit earlier. However if for some reason it can't be done by them in flame* I would usually use a tracked GARBAGE MASK (in KEYER) to generate the matte for that shadow.

Another example: if someone is standing in front of a table, his/her shadow will first hit the table (rather sharp shadow) and then the floor behind the table (more out-of focus). In such a case multiple mattes need to be used and cropped accordingly.

If, for instance the character is walking on a rough surface, the shadow absolutely needs to follow the "topography" of this floor. A technique that usually works is to render a flat-floor shadow (a matte) and then break and bend it according to the rough terrain in the warper. To do this just track the whole grid -both source (unmodified grid) and destination (bent grid) according to the floor's movements.

One subtle detail that many compositors tend to leave out is the shadow in front of the character. Look around you. Look at the table next to you. Notice that small blurred shadow below the table-leg, just before the floor? Even when a hard spotlight is hitting something and generating a sharp shadow behind it, usually there is a small distance between that object and the floor. A good example is shoes, whose edges don't generally touch the floor. So, if that floor can be seen, it can't be as well lit as the rest, for there's something blocking the light. (This is something that really helps blend 3D CGI into a scene.)

In most cases you will need to use GAUSSIAN blurs (the G button next to the layers-box in action) to achieve believable soft edges, and in some cases you will need different levels of blur at different distances from the shadow-casting character. This can be achieved creating different layers for each softness, then using the same matte for the shadow, and then blending these layers with very soft-edged GARBAGE MASKS in the KEYER.

When looking at a scene all these seemingly little details (that might by the way get very time-consuming) make us either buy it... or not.
 
note from Jeff Heusser:
I would add that shadows are not created in nature by taking a black frame and mixing it on top of the background! This method produces dull flat shadows with reduced detail (which can also be used as a "look'). I prefer to use color correction for most shadows, it is worth using a source node or pre-rendering your mattes to do this.
 
6- Other CGI (i.e. smoke, clouds, fog) [top]

If there are clouds/smoke/fog in the scene, it is highly important to make the character blend into that atmosphere which might involve creating some mattes with particles and then blurring them, (or using some other techniques), in order to have the character behind that "atmospheric" layer.
 
7- Focus [top]

By now you should know your scene well enough to be able to feel what part(s) of the CGI need to be blurred. Don't worry about blurring the main character or product shot if necessary. If the shot was shot properly, it will come into focus at some point (hopefully!). Once again, use only GAUSSIAN blurs. The extra render time IS worth it.
 
note from Jeff Heusser:
Often blurs are difficult to match using only the blurs built into flame*. Small point light sources (like stars) in a rack focus that need to be matched will usually require using sparks. I have had good results with Sapphire's RackDefocus spark.
 
8- Reflections [inicio]

Anyone who has ever done 3D at some point in his/her life knows how important reflections are to make evident what material something is made of. When doing shots where our character is in front or on top of a reflective surface we must remember to create an inverted (and if necessary deformed) copy of our image. The points to look at are transparency and color correction so that the reflection will seem to be part of the object reflecting it. And, of course, if the reflective object should move or rotate, the contained reflection moves in the opposite direction at a faster speed. Also, remember that if your client says that reflection is too evident you can always tell him that "objects in mirror are closer than they appear" (haha).
 
9- Grain [top]

Although for some footage with little grain the standard tip (95-98% transparent layer of color noise with itself as matte on top on the whole composite) usually works, if I have the time, I don't use it because:
- unless darkened it lightens the image a bit
- it gives even more grain to the film footage

So my work-around is as follows:
1- Generate a "noise matte" (monochrome) with shrunken edges (2-3 pixels) for the 3D.
2- Use the front (CGI) in 2 layers in action. First with its original matte, then with the noise matte we just created.
3- Color correct (MASTER LUMINANCE GAIN and MASTER HUE SHIFT) by turning these values up a bit in one layer and down a bit in the other (by the same
amount to keep the overall average). This color correction is additional to the previous one (step 3)

By doing this you can adjust the grain level to match the background without adding extra noise to the back.



Of course every preceding word will only work if our key is perfect. To get it we must avoid edges that are too visible or too  sharp. This is all done inside of the KEYER. It´s good to play with different keying methods (RGB, HLS, HLV, ...) until we find the best one. CURVES are also very useful, play with them until you know how they work (check out the "additive mix" tip in this page).

And most important of all: In order to look real, 3D images NEED to be destroyed. So, if you feel the motion blur rendered by the 3D artists isn't enough, use the inverse parenting technique described in the "motion blur on static objects" of this web-page. Use blur to give adequate depth of field. Match grain intensity. Move animation curves so they don't all start/end at the same time, just randomize them a bit manually once you have set your "correct" keyframes (huuuge tip).

I wish you fun and creativity in your composites and would love to hear from you if this "tutorial" (can it be called that?) has been as useful to you as some of the other tips have been to me.
 

Eric Schaechter
Senior Compositor
Ollin/Studio - Mexico City
www.ollin.com.mx
eschaechter@ollin.com.mx