More accurate depth and position channel for VFX and compositing
We need pixel perfect depth channel and position channel! The current depth and position is a little bit rough and inaccurate compared with beauty passes, as well as other renderers, which causes problem when you want to do something in the post.
Also, if the camera is inside a volume fog, depth and position will be completely destroyed as the pictures show. We truly need more samples for depth and position, or some method which can bypass volumetric computing.
5 Comments
My depth passes in Eevee don't look like that. You can't split the Z pass in Cycles to it's own pass, and disable the volume object for that pass?
We can do that. But for a very heavy scene, it will take dozens of seconds to load up the geometry alone, rendering a single or multiple viewlayer for data passes really cost a lot more time.
Yes, one sample sampling for the info passes is causing a lot of problem in compositing. I understand that this might be the correct mathematical way of doing it in some case, but it doesn't work wells with texture alphas, fog and motion blur. Why not give the user some options like in Octane. By default in Octane all info passes are rendered with 128 samples but you have the option to change the number of samples, you can also choose in you want distributed samples or non-distributed samples, with or without pixel filtering. This would be a priority for compositing inside Nuke and the adoption of Cycles in the VFX industry.
Agree. Haven't run into any case where volumetrics in data passes were ever useful.
Well… I updated my plugin, now it outputs pixel perfect depth and position pass by viewlayer override material. Here is the link github.com/RolandVyens/Industrial-AOV-Connector