A 3D object in a recorded scene is supposed to have part of the footage placed over it.
(think of a monitor screen that someone touches with their finger).
A somewhat similar object was sprayed green and used in the scene (green monitor screen).
Workflow is:
video recording of scene with person interacting with green-screened object
tracking of scene and placement of 3D object in place of green object
keying of person interacting with green-screened object
color-corrections, etc.
result of person interacting with 3D object
Most steps no problem, BUT here’s the thing: the edges where the green screen and live footage come together (an arm crossing the green screen) cause problems, the green screen gets jagged due to the keying mask used. I’ve been trying countless of masks/mattes combinations and options for something I expected to be simple… can’t get it corrected.
I haven’t been able to find any tutorial explaining this type of effect specifically. Most just deal with footage with background green screen effects or placement of single objects, not with object-based green screen (or whatever this would be called) where live footage partially covers the 3D…
Maybe someone can point me in the right direction…
How are you masking it? It shouldn’t look that bad.
Here is an example of what I think you want to do. It isn’t green screen, though the real world video camera post and the fence on the left are masked to appear in front of the rendered object. You can adjust the ‘feather’ for the mask:-
A good quality key (cut out shape) from green or blue screen footage is always difficult and depends upon how well you account for the technical issues.
Use the highest quality capture possible. Avoid highly compressed video.
Light the screen evenly to avoid heavy shadowing from the foreground (live) element.
Try to use a fast shutter to eliminate blurred edges on the foreground element. Failing that try edge lighting to accentuate the contrast between live element and green (or keying) screen.
All these problems need to be addressed otherwise you get issues with bad edges as Organic suggests. Then you have to down grade the key using feathering or the erode node.
Here’s an example of bad keying that I made:
And a quick demonstration of the steps required to get the key working:
Here is a key in action:
There are lots of tutorials for effective chroma keying on the web, all the techniques apply to Blender too. Just some of the tools vary in quality.
I’ve been able to get the keying correct, that wasn’t a problem.
I finally figured out where the composite is going wrong.
There is a masking problem with color correction that I can’t get solved.
Look at this image. Two simple masks, one color corrected. Why does this edge-coloring happen?
And is there another way of combining two masks with alpha channels and color correction?
Argggh, this is driving me nuts!
And to be clear: it doesn’t matter if ‘convert premultiply’ is used or not… And the color correction doesn’t matter either. Same problem when another image with flat color is used.
I would expect two alpha channels would just stay an alpha channel when combined in this manner.
I’m wondering if it isn’t a bug…
It actually uses the color info from the TOP image. If the source image of the bottom node would be orange, the edge would still be purple. For alpha transparency on top of the non-alpha part of the top image, this makes sense. But not for the entire mask.
That’s just strange… I’ll see what response I get from the bug tracker…
Ok, thanks to Sergey from Blender, I got this figured out.
Apparently this was done on purpose. The node path takes the alpha channel of the other mask into account.
After setting the Alpha channel, an Alpha Convert node is required to get ‘true’ alpha (for lack of a better term).
IMHO kind of counter intuitive, but it works now.
Eventually I managed to get a decent result of footage partially over a green screened object including fake shadows, specifically by NOT using the Keying node. Maybe I’ll do a tut some time…
