#Renderer & transparency
8 messages in this thread
Hi Steve,
Happy new year, I just wanted to let you know, that with Version 2.07 most of
our problems are gone.
The following in not ment to hurt anybodys feelings it is rather a wishlist and
from what I've read and heard about version 3 some of the things will probably
be included into the new version.
The main thing is the renderer. I'm not at all complaining about the speed
(well, it *could* be faster but there is *no* program around that is fast
enough). The problem is transparency. – Whew, Andrea just handed me the paper
with her notices, I hope I can translate that into english, so that it makes
sense to you:
The basic problem seems to be that transparency factors are *added*, where they
should probably be *averaged out*. Here is an example: a white, 80 %
transparent segment is placed partially in front of a black background and
partially in front of an opaque white object, which appeares dark grey, due to
its shadows. The intersections of the two segments should now look grey also,
even thou a little lighter. Fact is the intersection looks almost plain white.
A probably better example: a dark red (140/0/0) segment that is 70 %
transparent is placed in front of another dark red seg. (same color). This
second seg now is opaque. The color of the intersection should also now be dark
red, but it is bright red. The two reds seem to be added rather then averaged
out: 140 red + 140 red = 280 red. This is turned to the closest number possible
to display I presume: 255 (bright red). Instead the color-calculation should
probably be something like this: 30 % of 140 (the transparent seg) + 70 % of
140 (the opaque seg) = 100% of 140 red. I am not sure about that thought
Well I know that these examples are rather theoretical. The basic problem to me
seems the color-theory here. You seemed to have used the additive color-mixing
theory instead of the substractive-mixing. A good example for the unusabillity
of additve color mixing might be my most gross red 70's ashtray. If I for
instance put that monstruos deep red glass ashtray onto a green underground the
green does not turn lighter in brightness – as in addtive color-mixing – it
turns darker, turns more dirty, towards olive – just as in substractive color
mixing.
This is easily illustrated by black & white photography. Lets say you want to
shoot a picture of a red brick-house that is partualy covered by ivy. If you
use a red filter the bricks become very bright, they will apeare allmost white
on the b/w print, wheras the ivy will turn allmost black. If you use a green
gel on the lens, the ivy now will be allmost white and the brick-wall will
appeare a lot darker then it actually is.
In conclusion I think the only problem is the wrong colormode. Additive
color-mixing does not work, because there is hardly a case where this is used
in real life (hey, of course, on monitors). Most of the times you see
substractive color-mixing. I can imagin that this problem sounds a lot more
trivial then it is to program. But I still wanted to bring it to your attention
just because we had some problems with a production recently.
Another thing was the reflection on transparent segments. If you have a
transparent white seg in front of a black background there is no reflection
rendered, even if it is cranked all the way up.
Wishes: Would it be possible to predefine motion paths in Sculpture, so that
they could be used as such in direction? That would be a big help if you have
characters move along mountains or so.
What do you think, is there a possibility to include motion paths in Action
too? On many occasions that would help to make the movements more
understandable, especially if you have different figures interact.
Would it be possible to have measurements all through the modules? For instance
that would be great if you build a movement of a person walking in action. If
you then import that movement into direction you have to somehow coordinate to
relative and the absolute motion so the figure does not look as if it is
sliding over the floor. If you would have measurements or a grid in Action that
uses the same units it would be a lot easier to get the figure moving according
to the underground in Direction. Also numeric entering of values would be
great. I know that this is possible even now, but unless you enter the value in
the channels the values are never exact (I did a movement (version 2.07) once
where I used "new transition" to start another movement with the same frame and
values. Unfortunately I errouniously changed the values on some segs (frame 1)
after that. I tried to just compare the values in the display and entered the
values according the the original ones. After I rendered both motions there was
still a big gab between the to movements. Comparing the values in the channels
showed the problem. The values were not identical. ass soon as I changed the
values to the exact ones everything was fine.)
Do you think an EPS import filter would be possible to include into Sculpture?
EPSes are bezier-splines as you know. Sculpture is a based on splines too.
Wouldn't it be possible to build an import filter for EPSF? I agree that dxf
files are cumbersome and in most cases it makes no sense using them in A:M. If
I need to do a logoanimation I usually have to build the whole logo in
Sculpture which is time consuming compared to just entering the text via
keyboard. I know that A:M is not meant to be a Tool for Logoanimation but the
control you have over the animation makes you want to use A:M for that kind of
stuff too. (I personally do not like flying logos a whole lot, but sometimes
you just have to do them to get a job) Dxf import does not allways work too
good (depending on what program you use to generate the dxf file) and besides,
you have to carry unnessesary masses of data all the way though the animation.
EPSes would probably be a more intelligent and elegant way to import logos
etc.. Of course this probably sound more trivial then it is too.
I wish you a happy 95 and again, the above is not meant to hurt your feelings,
it is supposed to help.
Peter
>>>>a dark red (140/0/0) segment that is 70 % transparent is placed in front of
another dark red seg. (same color). This second seg now is opaque. The color of
the intersection should also now be dark red, but it is bright red. The two
reds seem to be added rather then averaged out: 140 red + 140 red = 280 red.
Peter,
I have had a lot of similar problems with the renderer, some of them I was able
to go around… My experience is that when you have a transparent object you
have to pay *A LOT* of attention to AMBIANCE values of your transparent
object… Value "0" is your best shot… If you want the objects behind a
little brighter try "1" or "2"… Anything higher will give you the results you
describe…
I get excellent transparency from A:M… with the right settings of Mirror,
Specularity and Regraction it can beat any renderer on the market… Good luck!
Pawel
Hi Pawel,
(I've been reading your and Jim Jonah's messages to other users for some time
now and boy,I wish I'd new A:M as good as you guys do)
well, you are right about the render quality. You *can* get highly realistic
textures if you need them – if you have the time to spend. We *did* experiment
with ambiance a lot, still it didn't work correct. Our particular problem here
was that we needed a stylized look for the animation of a syringe but at the
same time it had to look *right*, not realistic but right. On the places where
more than one layer of tranparent material laps over other pieces the effect
appeared. B.T.W. the ambiance on all the models was set to zero.
I mean, I love A:M and I love working with it, but some things just drive me
nutso. (Just yesterday I accidentially erased a model only because Character on
the Mac can save Figures with .seg endings. I hit the save shortcut after I
imported the segment into Character, forgetting that this would erase all the
segment-data)
Another thing. Maybe you have an Idea. I am looking for a good "fog" material,
theoretically it is possible to do that, but do to the problem described
earlier all results look strange and not at all naturalistic. Any idea? I can
send you the data of a fog material we have allready build. It looks beautyfull
until you place an object behind it.
talk to you,
Peter
Hanspeter,
PMFJI, I missed the beginning of this thread, but from your last message it
sounds like you had problems with overlapping transparent objects. Have you
tried using the cookie cutter and the transparency maps to eliminate unwanted
parts of the transparent object?
On the fog issue, again have you tried using transparency maps and ambience
maps? Try using the same map in both places with the amount set to 100%. By
adjusting the gray values in the map, you can get some pretty cool effects. I
picked this tip up from TAnimate over an AOL, and have been experimenting with
"visible" lights. So far, the results have been very good!
I hope this helps,
Jim Jonah
Hi Jim,
Here is our basic problem again:
"The basic problem seems to be that transparency factors are *added*, where
they should probably be *averaged out*. Here is an example: a white, 80 %
transparent segment is placed partially in front of a black background and
partially in front of an opaque white object, which appeares dark grey, due to
its shadows. The intersections of the two segments should now look grey also,
even thou a little lighter. Fact is the intersection looks almost plain white.
A probably better example: a dark red (140/0/0) segment that is 70 %
transparent is placed in front of another dark red seg. (same color). This
second seg now is opaque. The color of the intersection should also now be dark
red, but it is bright red. The two reds seem to be added rather then averaged
out: 140 red + 140 red = 280 red. This is turned to the closest number possible
to display I presume: 255 (bright red). Instead the color-calculation should
probably be something like this: 30 % of 140 (the transparent seg) + 70 % of
140 (the opaque seg) = 100% of 140 red. I am not sure about that thought
Well I know that these examples are rather theoretical. The basic problem to me
seems the color-theory here. You seemed to have used the additive color-mixing
theory instead of the substractive-mixing. A good example for the unusabillity
of additve color mixing might be my most gross red 70's ashtray. If I for
instance put that monstruos deep red glass ashtray onto a green underground the
green does not turn lighter in brightness – as in addtive color-mixing – it
turns darker, turns more dirty, towards olive – just as in substractive color
mixing.
This is easily illustrated by black & white photography. Lets say you want to
shoot a picture of a red brick-house that is partualy covered by ivy. If you
use a red filter the bricks become very bright, they will apeare allmost white
on the b/w print, wheras the ivy will turn allmost black. If you use a green
gel on the lens, the ivy now will be allmost white and the brick-wall will
appeare a lot darker then it actually is.
In conclusion I think the only problem is the wrong colormode. Additive
color-mixing does not work, because there is hardly a case where this is used
in real life (hey, of course, on monitors). Most of the times you see
substractive color-mixing. I can imagin that this problem sounds a lot more
trivial then it is to program. But I still wanted to bring it to your attention
just because we had some problems with a production recently.
Another thing was the reflection on transparent segments. If you have a
transparent white seg in front of a black background there is no reflection
rendered, even if it is cranked all the way up."
That was the first message.
As you see, it is *not* a problem of removing parts of verlaping segs, its
merely a problem of showing them but not in solid white. The actual problem
occured on a scientiffic animation we did on eye-surgery (I know, some jobs you
have to do). First a special type of syringe. That things had two tubes. Each
of the tubes contains one component of a special fibrine glue. The animation
had to look real, but schematic at the same time. Too much realism would have
given it a too scary look. The way we have it now is far too schematic though,
it doesn't look right. The clear cornea was an even bigger problem, since there
was supposed to be an incision shown. That meant a *lot* of muscle motion and
overlapping transparent pieces all over the place. We ended up doing most of
the stuff in post.
Thanks on the hint for the fog I'll give it a try, especially a rotoscoped
black and white QT-Movie would probably work great for that. Any experience
with that?
Read ya,
Peter.
>>>I mean, I love A:M and I love working with it, but some things just drive me
nutso.
Peter,
this seems to be the standard relationship with A:M ;=)
The transparency troubles you describe are just part of those "quirks"
everybody sooner or later runs into… The reason I jumped into the tread was
because I have been there (overlapping bright areas) before and with careful
adjustments I got rid of them… I use A:M mostly for a high resolution product
still shots, including many transparent cosmetic containers with color gels,
foil stamps printing and other tricky objects, including whole bitmap
background scenes shown through infracted transparent shapes. (I am a
commercial artist, with about 80% of my business in packaging design). Because
of this kind of work I have spent huge amounts of time playing with the
transparency issues… I hear what you are saying about the additive and
subtractive color model, and can't really judge if this is the cause of our
trouble or not. My experience has been to pay extremely close attention to
ambiance when dealing with the transparency… (my guess is that the bug, if
there is such is in this area rather than the color model) If your transparent
object has the Ambiance value 0, it is possible that the object behind is "too
ambient", or "not ambient enough" and in order to compensate the scene has
light sources that are too many or too strong …
As for the "fog" material, the only thing I've played with was an attempt to
create a "glow" around certain object by placing a simple 4 point patch with
transparency mapping (specularity was 0, to avoid highlight) — I almost got it
working except for the edge that has rendered as totally opaque — it goes back
to this dreadful "background color" that shows up when you need it the least…
Since the "glow" area is supposed to disappear into the scene, the grayscale
bitmap used for mapping starts black out of the frame and gradually gets
lighter… Well, the black becomes the "background color" (so handy in cookie
mapping!) and instead of being 100% transparent, becomes 100% "nonexistent" so
the segment can fully show itself <BG>
If for some reason you do not have this problem in the Mac (it certainly is a
bug — just appeared to me that it might even be file format specific, let say
TGA or PCX — I will have to test it), you might want to try the transparency
mapping to create fogs, clouds, glows and so on… Since transparency mapping
is not the same as transparency in A:M (it is described as simple color
averaging) it shouldn't give you any trouble with the ambiance… — Hey, why
don't you try to wrap a tiny (10×10 pixels) gray bitmap on your syringe as
transparency mapping and try to render it that way… It looks like crazy idea
but it might work. (Unless the "background color creaps in <g>).
Pawel,
I just tried a trick I learned over on AOL using the same map for transparency
and ambiance to create a visible light source. The map was a gradation from a
light gray to black, saved as a Pict file, and everything worked perfectly for
me. Your "black/background" color problem may well have been a file format
problem. When I used the same file saved as a Targa, there was a strange white
ring that showed up about 1/2" past the edge of the gradation. That disappear
when I resaved the file as a Pict format.
Hope this helps,
Jim Jonah
Thanks for your input Peter! We always appreciate hearing from our users as
this is where we get the info we need to improve our product…. hopefully we
will be able to make some of your wishes come true with version 3!
Steve, Hash Inc.