2.5d ReLighting in Fusion; CustomTool (.setting)

[zarti]

hi to all of you,

spend two days to find the 'trick' to use world-position info in Fu' ...
fortunately, someone else has done it (i cannot remember "who" and "where", maybe Pigsfly or Vfxtalk) and i thought i could share it here with you.  =)

1- load an rendered image which has p-world data
    *if you need an image to test, download this one (EXR file; a simple lightwave scene rendered with shaderMeister)

2- load a CustomTool, then load the ".setting" file attached here.

3- connect them both (see below)

4- have fun!  =)



++++
below i'm posting some screenshots from my first tests.


hope you find it useful, (of course, if you actually do not have it).   

[MrWyatt]

Hi zarti. nice to have you here. 
I have made pretty much the same thing in nuke here : http://forums.db-w.com/index.php/topic,450.0.html
I just used it to make stereoscopic proof masks, but I guess one could also do a fair bit of relighting using it. I see you have the same problems as I have with borders though. I guess it would help to render the pworld pass in double resolution to get rid of a few hard borders.
Thanks for the pictures. will see if I can replicate he expressions in nuke, as my mask is not falling of so nicely spherical as yours.

anyway. really nice tool.

[zarti]

yea, i'm happy too to meet you guys here again! 

I have made pretty much the same thing in nuke here : http://forums.db-w.com/index.php/topic,450.0.html

in fact, your posts provoked my search to achieve the same thing in Fusion. so, thank you!     
 

I just used it to make stereoscopic proof masks, but I guess one could also do a fair bit of relighting using it. I see you have the same problems as I have with borders though. I guess it would help to render the pworld pass in double resolution to get rid of a few hard borders.

maybe not only for relighting; you can easily extract the data for one of the axis and use it for fog and interesting other post-effects.


regarding hard borders; i tried a lot cameras and camera settings and the only render which behaves correctly is that which has no AntiAliasing active. (see attached pic below)
the AA is the problem and this seems to be a LW-s issue. i don't know if it is worthy to post that like a bug on NT forums. i do not know either if other 3d packages have/generate the same problem ... 

Thanks for the pictures. will see if I can replicate he expressions in nuke, as my mask is not falling of so nicely spherical as yours.

anyway. really nice tool.

well, notify me on your progress.    oh ... and again, i'm not the maker of that, just the transporter.   

[lightwolf]

regarding hard borders; i tried a lot cameras and camera settings and the only render which behaves correctly is that which has no AntiAliasing active. (see attached pic below)
the AA is the problem and this seems to be a LW-s issue. i don't know if it is worthy to post that like a bug on NT forums. i do not know either if other 3d packages have/generate the same problem ... 

I'd say it's a general problem that comes with AA.
Ideally one would want to composite and then perform AA - which is practically what rendering at a double (or higher) resolution does. One could of course try to design a file format that stores more than one sample per pixel to allow for AA in post. But, again, that would result in massive size files (i.e. LWs adaptive sampling can easily create hundred samples per final pixel - which, in an image file, would be the equivalent of 100s of extra RGBA data stored per final image pixel).

The Coverage buffer that 3DSMax exports (combined with a background colour buffer) is designed to get around the issues, but even those only do so to a certain degree.

The only package that I know of that actually handles these issues correctly is Blender, as it can composite before AA. On the other hand, that also means that you need to re-render the 3D part of the comp every time.

Cheers,
Mike

[zarti]

i think i found something ...

totally doable inside the compositor (Fusion). see two different moments below.
they are zoomed at 400%.




maybe it is not scientifically correct but i like the results.
my Fu' keeps crashing and project keeps being corrupted because (maybe) my laptop @home, where i'm making tests, is 5 years old.
but fortunately i can open and correct it in a text editor. hope all those worth it.   =)

[zarti]

some other moments ...