"Screen" vs. "Add" transfer mode. Which do you use, and why? Discuss.

"Screen" and "Add" are the photoshop names for a couple of modes of adding images.

Add is add. Simple really. Adding two images "a" and "b" to get the "out" image the "equation" would be:

Rout = Ra + Rb
Gout = Ga + Gb
Bout = Ba + Bb

Screen is the inverse multiplication, i.e. assuming the range of RGB values are 0.0 to 1.0 you get

Rout = 1.0 - ((1.0 - Ra) * (1.0 - Rb))
Gout = 1.0 - ((1.0 - Ga) * (1.0 - Gb))
Bout = 1.0 - ((1.0 - Ba) * (1.0 - Bb))

This "sorta simulates" the action of physically overlaying two negatives and-re-shooting that...

Disregarding the math, we all know how they "look". I find neither works well.


Technically, light-effects are inherently additive. (A glow can never block light behind it, can it?!) So you "should" use "ADD" all the time.

Those of us who has tried to us ADD, will soon find it becomes a bleached burn-out mess. This is because the ADD happens in linear pixel space, which isn't truly linear as light levels go (and camera respons aint linear either).

So many people use "Screen", but Screen is really just a hack to get around the problems of Add vs. the nonlinearities. Screen can easily generate a *too dark looking* glow. I hate "dark glows".

It totally ruins the effect, for me, to see someone whip a lightsaber over, say, a bright sky, only to see the red glow of their sabre appear *darker* than the sky appears. *argh* (Broken Allegiance, Dark Skies, most films have sequences where the glow is too dark and actually darkens the background).


My questions to you guys are

1. Do you guys agree? Or did I overdoo those mushrooms?

2. Which do you generally use? Screen? Add? Why?

3. Would you guys want another, better, more "light/energy effect looking" transfer mode?

4. You want me to shut up already ?

/Z

 

Well, for me it all depends. I simply try both on my shot and see what looks good. I think Add makes for a "hotter" look but screen is a lot "cleaner". For ALD I mostly used Add because the background was fairly even most of the time and the "messyness" of the glow helps disguide sloppyness in rotoscoping.

On the other hand, for the forest fights in The Formula, I used almost completely screen mode since Add mode bonded to, and really exploded bright patches in the backround and made it look more blobby.

So, again, I think it really comes down to the shot and which transfer mode works best with your background.

 

The word of the lord...

Now let us reflect in a moment of silence.

 

The technicality of that post is amazing, I had to read it twice to understand all of it. I myself use the Screen transfer mode most of the time. And from #3 at the bottom of your post I get the impression that you may be willing to develop another transfer mode for us. Or did I misunderstand that?

Regards,
Kai

 

The thought occured to me, yeah. Although it wasn't the main reason for doing the post... main reason was curiosity on what is used.

As for developing one.... the question is.... is it worth the effort?

(Btw, thanks for reminding me of how "add" tends to magnify background things, kinda forgot that.... again a side-effect of the adding in 'pixel space')

/Z

 

I personally alternate between using the screen transfer mode and using this pluging called "knoll unmult" its a free plugin designed for establishing alpha channels and such, or something like that. It tends to preserve the color a bit more and looks good in some instances, I mostly use it to help preservere that deep red of vader's and dooku's saber thats pretty hard to get unless the lighting is very dim.
Any way, give it a try I think you can find it at Puffin design's website, or do a search for it. Sorry I dont have a link around

 

I tend to use screen for sabers. For a more surreal, "hot" look, as Ryan mentioned, I use add. The problem is it means that any colors near white end up blowing out to white. It has its place, I used it for a number of things in the Formula, particularly the "Vision" fight.

In the mirror fight, however, I used screen almost exclusively. I find that it, not add, actually produces a more pleasing, and simultaneously cinema-realistic, effect, in terms at least of saber glows.

It doesn't obscure glows from behind, as you'll notice. Even though it's not the direct addition, the inverse multiplication will still make white come out white. So it reacts realistically to light in front and behind. Red and green glows crossed will make yellow, as they do in reality.

I go by the theory that a lightsaber is actually a brightly-colored energy rod with no surrounding haze, were one to actually exist and be viewable in person. Like a welding iron. The glow aura is an artifact of the lightsource being so bright that it blows out the film/video to white and leaves that "glow" as a residue (much as a filmed welding iron will blow out to white with a blue haze).

Hence, the aura itself does not actually add to the surroundings, as it is not technically there.

ALSO, in response to the following:

It totally ruins the effect, for me, to see someone whip a lightsaber over, say, a bright sky, only to see the red glow of their sabre appear *darker* than the sky appears. *argh* (Broken Allegiance, Dark Skies, most films have sequences where the glow is too dark and actually darkens the background).

I think you may find that in these, and other such cases (AQOF is another, and Rogue as I recall but it's been a while), they used neither add nor screen mode to insert the glows, but something (possibly even "Normal") specifically utilized to make the glows visible.

I don't subscribe to this way of doing things, but the reasoning is that if you're fighting in front of a bright sky, the sabers (if obeying the laws of light--which they violate by nature but we won't get that Trekkie about it) are going to disappear, and you won't be able to see the fight.

So they make sure you can see the sabers, even though I've never seen it look good like that.

M. Scott

 

Confession time. When I did the rotoscoping in Dark Skies I didn't even know about the screen and add functions until half way through, so I just had them set to 'normal' most of the time

Now I use a combination of either screen or add (or whatever transfer mode looks cool) but, by and large I would say I use screen, since I prefer a softer, more subtle glow. But this is just my own personal preference.

 

^Told you.^

M. Scott

 

This last summer I shot some footage in the desert and rotoscoped it. The background was fairly bright, so i was messing around with different transfer modes on different layers and i found that for certain colors, using add and screen in different glow layers really helped make it brighter...however i cant remember exactly which layers used which, but regardles, try mixing the two and see what you get....

 

So many people use "Screen", but Screen is really just a hack to get around the problems of Add vs. the nonlinearities. Screen can easily generate a *too dark looking* glow. I hate "dark glows".

I think you misunderstand the purpose of Screen, it's certainly not a hack.

The difference is in the use. Add is when you're compositing the light itself onto an image. Screen is when you're compositing an element that is already correctly rendered, but needs to be integrated.

What screen does is correctly simulates what happens with over-exposure. If you had a strong light in a scene, the other objects would darken on film, because of the dynamic range. This is what screen does. You need this when you have separately rendered lit elements.

If you render elements such as lens flare in CG separately, you would use Screen to composite them. Glows are inherently something that get added onto, unless they are CG glows rendered as a pass; in that case, since the image is already 'correct' and needs only to be integrated, screen would be used.

Again, here you have to be careful with the meaning of 'glow', as to whether you mean an atmospheric effect or a lens effect.

 

One thing I thought of today: In, say, Geonosis, it doesn't look like either method is used. I've tried to reproduce the sabers there, and no-go thus far. Also, in RotJ, Luke's green saber blade is VERY visible against the sand. Here's an idea I haven't tried yet.

Create your roto layer (AE only, sorry) and make white shapes WITHOUT adding a black background. do the blades as Ryan_W says, compositing ONLY THE BLADES together with "Add" or "Screen." Then, composite the BLADES w/o a black background onto the footage using NORMAL transfer. Or, if you want it to respond to the background, try duplicates that aren't at full opacity, like a copy at 50% on normal and a copy at 50% on add.

 

Here's what Adobe Photoshop 5.0's Help pages stated about the two in simple english.
Add

Adds the pixel values in two channels. This is a good way to combine non-overlapping images in two channels.

Screen

Looks at each channel?s color information and multiplies the inverse of the blend and base colors. The result color is always a lighter color. Screening with black leaves the color unchanged. Screening with white produces white. The effect is similar to projecting multiple photographic slides on top of each other.


That being said...screen would be the closer choice for trying to recreate film effects back in the day sorta thing...like the bit duscussing the nasty matte lines in Empire Strikes Back for instance...how they composited them basically. (Been mentioned already...O.K. )

BUT, Add does exactly that...it works strictly by adding the two which means you'll have area of brighter pic due to the 2 images overlapping. For darker/coloured explosions (especially in space scenes), brighter, bigger stars might turn parts of let's say a blue explosion pure white...ARGHHH!

Screen is a little safer way to go...since its closer to masking with an Alpha Channel. Problem is parts that are 'pure white' are the only parts that'll ever be 100% opaque...everything else will be transparent..so purple lightsabres would be less visible (or stand out less) than say yellow lightsabres (which are lighter and thus more opaque) due to their repsective positions on the colour range.
It's just a matter of what looks best.

 

DaftMaul: *normal* (aargh, the pain, the agony *grin*)

I go by the theory that a lightsaber is actually a brightly-colored energy rod with no surrounding haze, were one to actually exist and be viewable in person. Like a welding iron. The glow aura is an artifact of the lightsource being so bright that it blows out the film/video to white and leaves that "glow" as a residue (much as a filmed welding iron will blow out to white with a blue haze).

EXACTLY. The "haze" you get from welding (not counting the smoke also produced, but the "glow" around welding) is the optics inside the camera bouncing light around, the fact a single piece of film is lit so brightly it lights it's surroundings!

I.e. precicely as you say, the "white rod with red glow" is in 'reality' a red super bright rod, overexposed to white w. a red haze as the result.

There is also this little detail of "light level". Light is highly non-linear. Our eyes dynamic range is astounding. So the sabre glow shuld *really* be very different depending on the lighting situation. In an outdoor scene you shouldn't really see much of it at all!!!. (Shoot a little outdoor sunlit welding vs. midnight welding and see the difference). I think all movies and fanfilms totally neglect this lil' detail (But hey, it's "art")

I don't subscribe to this way of doing things, but the reasoning is that if you're fighting in front of a bright sky, the sabers (if obeying the laws of light--which they violate by nature but we won't get that Trekkie about it) are going to disappear, and you won't be able to see the fight

I dont subscribe to it either.... the thing is us humans see stuff *remarkably* well when it moves, even though its subtle. When one sits infront of ones computer and can hardly see the sabre against the sky in the *still frame* one is watching, and begin mucking about w. transfer-modes to "fix it", I think one is making a mistake.... first see if it is really 'gone' even in a moving frame, chances are it isn't.

And if it still is, then make the sky 2% darker then I mean, whats wrong with dark skies

I think you misunderstand the purpose of Screen, it's certainly not a hack.

The difference is in the use. Add is when you're compositing the light itself onto an image. Screen is when you're compositing an element that is already correctly rendered, but needs to be integrated.

As I understand it, "Screen" is indeed a hack in a mathematical sense, i.e. doesnt simulate *anything* real world except possibly overlaying two negatives and shooting the result. I mean thats not a real-world thing, it's a special-effects thing in itself.


I may be wrong though, because you then write:

What screen does is correctly simulates what happens with over-exposure. If you had a strong light in a scene, the other objects would darken on film, because of the dynamic range.

Explain this. You have some hard data on this? What in film causes this to happen?

/Z

 

you then write:
What screen does is correctly simulates what happens with over-exposure. If you had a strong light in a scene, the other objects would darken on film, because of the dynamic range.

Explain this. You have some hard data on this? What in film causes this to happen?

Screen is used when you're compositing a light element that does not hide what is behind, like a lens flare or a lazer effect. So when you composite these, you have to think about what will happen to the objects you can see in these semi-transparent areas.

What would happen if you were to film this scene with a camera? Well, there would be more light in the scene, so in order to keep correct balance, you would limit the exposure. If you do that, the darker elements will become even darker.

One could explain this with numbers. Let's say a scene has an amount of light that is 100 of some units. A table in the scene is receiving 10% of that light : 10 units of 100 units. Now you composite another layer with objects that add another 50 unit of light, for a total of 150 units. If you put that back on 100 units, the table is now proportionnaly darker.

If you weren't doing this remapping back to 100 units, the image would just clip all the brighter values and it would look aweful.

This is all necessary because in reality the sun is one millions time brighter than a white t-shirt, and film or digital images, or the eyes, do not have the range to show us all of these values. So we're always remapping a black point and a white point that maximizes the range of colors that are useful without loosing too many low end detail.

The Add mode works on images that haven't been exposed to their optimal range. If they were, adding any more color would cause clipping.

 

I read what you write, and you are actually saying exactly the same thing I am saying, i.e. that light is very non-linear.

However, you are not really defending the "Screen" mode's behaviour in doing so. At all. You write:

One could explain this with numbers. Let's say a scene has an amount of light that is 100 of some units. A table in the scene is receiving 10% of that light : 10 units of 100 units. Now you composite another layer with objects that add another 50 unit of light, for a total of 150 units. If you put that back on 100 units, the table is now proportionnaly darker.

Well yes but this doesnt happen all over the image, which a "change in exposure" would do, it ONLY happens in the area where there is semi-bright in the supermiposed image. Any blackness in the superimposed image creates NO CHANGE in the original image, at all.

So it doesnt follow your "it simulates exposure" at all.

What it DOES simulate (poorly) is non-linear addition. Sorta like when you think a 20 dB sound and another 20 dB sound would make a 40 dB sound. It doesn't, it makes a 26 dB sound, because the dB scale is logarithmic and 6dB up means "double amplitude".

If you weren't doing this remapping back to 100 units, the image would just clip all the brighter values and it would look aweful.

Blow outs (clipping) is very useful and doesnt look "awful" at all when done right. The entire core of a lightsabre is a blowout (or is representing one, anyway)

I re-iterate: The "screen" transfer mode DOES NOT simulate anything that ever happens inside a camera (it is useful for explosions which is gaseous particles which can actually block light behind them). The only thing it approximates is overlaying two negatives (which in itself is a special fx technique, i.e. not an "in camera" event).

We need nonlinear addition. Screen is a hack.

In my old raytracing software RayTracker all light was added non-linearily, and it made a hell of a difference in image realism. I have still not seen any software on the market who does this, and that was 10 years ago I did it. Hmmm.

I am just now looking for a job at ILM to get this fixed once and for all (Yeah, jobless right now, annoying, innit?)

/Z

 

i just use add, unmult, and screen how I see fit. Go for what looks best, or fits your personal preferance!

 

You are correct, MasterZap