Tweet
I originally put this in "issues"--but apparently that's for bug reports--
I'm rendering a moderately simple scene, three soft deforming objects squeezing into a fourth. The whole scene is about 360 frames. The objects are SSS. The problem is that about 1/10th of the frames render with this canvas-looking pattern on the objects--it looks like some sort of error in how the images are processed. You can see this attached--the one corner of the frame looks fine, the next frame it is checkered and has weird patterns (these are just closeups of the lower-right corner of the image). I ran it again with a simple non-SSS material with no image maps attached, and you can see more dramatically what is happening on the bum frames (and these are the entire frame), as the whole object more or less disappears. (It is also probably worth mentioning that the render times on the bad frames is more than twice as long as the good frames.)
The objects don't seem to do this with Karma, so it appears to be a V-ray issue (maybe? Or probably user error). There is nothing strange going on with normals or unruly point velocities (that I can see--that was the first thing I checked). The scene is a lot of lovely quads, no n-gons in sight. The bad frames occur consistently throughout, but not based on any change in action--the two frames I'm focusing on here are adjacent and one is bad, the other is good. Also, it happens to the four different objects at different times, so it's apparently not a frame-wide issue. The shot is pretty consistent in its setup--one simple rectangle light, no moving camera.
Given that I've already spent 30 hours rendering the good frames, I'm hoping this is an easy fix!
I'm on Houdini 19.0.657, tried with the latest nightly build of V-ray with the same result. I also tried using 19.5 and the latest V-ray build, but still got the same result. And I'm running a rather capable Windows 10 machine.
Thanks for any help!
ADDENDUM:
After using some mild profanity I figured I would just save everything as alembics and try to get something good out of UE4. Loading in one of the alembics to my scene and rendering from that, rather than the cached geo (which was the exact source of the alembic), and the problem appears to go away, so is there just something strange about how V-ray deals with cached geo?
ADDENDUM TO THE ADDENDUM:
Rendering using the alembics instead of the .bgeo cache appeared to work, except the multimatte did not work--the 0 object IDs rendered red with the 1s. Re-insert profanity here.
ADDENDUM TO THE ADDENDUM TO THE ADDENDUM:
Looks like part of the problem is with the subdiv. By using the alembics, then unpacking them, I got the beginnings of a result. By then subdividing I was able to get the motion blur and get rid of some, but not all, of the jaggies--the spec channel still required a lot of blurring in Nuke to get rid of the faceting. I noticed that using a subdivide in the geo node gives better results than using the V-ray subdiv, so I'm guessing my workflow (coming out of ZBrush) might be to blame--but regardless, the results are pretty unusable, leading back to Mantra. Help me help me!
I'm rendering a moderately simple scene, three soft deforming objects squeezing into a fourth. The whole scene is about 360 frames. The objects are SSS. The problem is that about 1/10th of the frames render with this canvas-looking pattern on the objects--it looks like some sort of error in how the images are processed. You can see this attached--the one corner of the frame looks fine, the next frame it is checkered and has weird patterns (these are just closeups of the lower-right corner of the image). I ran it again with a simple non-SSS material with no image maps attached, and you can see more dramatically what is happening on the bum frames (and these are the entire frame), as the whole object more or less disappears. (It is also probably worth mentioning that the render times on the bad frames is more than twice as long as the good frames.)
The objects don't seem to do this with Karma, so it appears to be a V-ray issue (maybe? Or probably user error). There is nothing strange going on with normals or unruly point velocities (that I can see--that was the first thing I checked). The scene is a lot of lovely quads, no n-gons in sight. The bad frames occur consistently throughout, but not based on any change in action--the two frames I'm focusing on here are adjacent and one is bad, the other is good. Also, it happens to the four different objects at different times, so it's apparently not a frame-wide issue. The shot is pretty consistent in its setup--one simple rectangle light, no moving camera.
Given that I've already spent 30 hours rendering the good frames, I'm hoping this is an easy fix!
I'm on Houdini 19.0.657, tried with the latest nightly build of V-ray with the same result. I also tried using 19.5 and the latest V-ray build, but still got the same result. And I'm running a rather capable Windows 10 machine.
Thanks for any help!
ADDENDUM:
After using some mild profanity I figured I would just save everything as alembics and try to get something good out of UE4. Loading in one of the alembics to my scene and rendering from that, rather than the cached geo (which was the exact source of the alembic), and the problem appears to go away, so is there just something strange about how V-ray deals with cached geo?
ADDENDUM TO THE ADDENDUM:
Rendering using the alembics instead of the .bgeo cache appeared to work, except the multimatte did not work--the 0 object IDs rendered red with the 1s. Re-insert profanity here.
ADDENDUM TO THE ADDENDUM TO THE ADDENDUM:
Looks like part of the problem is with the subdiv. By using the alembics, then unpacking them, I got the beginnings of a result. By then subdividing I was able to get the motion blur and get rid of some, but not all, of the jaggies--the spec channel still required a lot of blurring in Nuke to get rid of the faceting. I noticed that using a subdivide in the geo node gives better results than using the V-ray subdiv, so I'm guessing my workflow (coming out of ZBrush) might be to blame--but regardless, the results are pretty unusable, leading back to Mantra. Help me help me!
Attached Files
Comment