
Originally Posted by
Wil Renczes
Johnny:
As an engineer that works on the product, I can tell you for a fact that I've never heard of a bug around shadow/highlight. We're certainly not sitting on a known bug (that thread you referred back to is a user thread, and they're talking about auto levels changing with the footage dynamically & potentially creating a 'pumping' effect if there are lighting changes in a given shot, hence the suggestion to use temporal smoothing options when using auto levels). The tearing in your screenshots is something else altogether - i suspect that is a bug, and we need to repro it. I think you touched on what is probably the key issue, which is the consecutive clips each having the effect. Have you noticed whether it's specific to having GPU acceleration enabled, or are you seeing this in software mode as well?
On a related note, you've been complaining a lot about render performance, and touched on the Performance vs Memory thing. Unless you have a specific requirement for Memory mode, don't use it. What it does is reduce the amount of render threads running, which in turn lowers the amount of memory used to compute intermediate frames. This was more of concern back in CS4 & earlier, when the app was still a 32 bit process. At this point, I think it's pretty much an obsolete option and should be left to the default of Performance, unless you're running with 6 or less gigs of ram, which is just a bad/pointless idea these days.
Also, while looking at your posts on the Adobe forum, I couldn't help but notice that you have something like 6 or more instances of Colorista applied to a clip. This is probably what is bringing you render times to its knees - each layer has to be composited to a flat RGB (or possibly YUV if they do native YUV effects, although I suspect not), then the effect is applied, then the composited version is passed again to another plugin instance, recomposited again, times 6. Not optimal. I don't know how it generally performs on 4K material, but with six layers of their plugin on a given clip, you're probably breaking the bank right there.
OTOH, if you were using something like the native RGB curves / Three way color corrector / fast color corrector filters (heck, all three) - in CUDA, these are computationally combined on the GPU and happen immediately. I think you'll save yourself a ton of render time by doing whatever basic grades you can via GPU, and if necessary, try to sparingly use one or two levels of Colorista on the topmost layers.
HTH, Cheers