I have been doing all my tests, and having lots of fun, but find a bit of a hitch moving between editorial renders and my original 4K footage. I am testing out some scenarios where I feel the Roto-brush tool will help nicely. I make the adjustments on the editorial proxy renders, and am happy with the outcome (I do realize that in any footage replacment there will be a need for some after-the-fact fix ups) untill I replcae the footage with the corrcetly scaled 4K originals. I notice that AE then has to reconfigure each frame with the new footage. If I want to render out the proxies to use as Alpha mattes, what format should I render them to?
Do I still need high quiality intermediate codecs for ALpha channel only, especially since edge detail is important, or is there a suggested Alpha codec to use?
I would use 16bit greyscale TIFF sequences. It's efficient in terms of file size and maintains a maximum of fidelity. QT Animation also has a dedicated greyscale/Alpha only mode, if 8bit is enough...
When would 8bit not be enough for a dedicated alpha channel?
I had the idea that using the alpha along with the 4K footage would save processing time trying to add the masking effect to the 4K footage itself - using whatever means, I assume that less "effects" = faster work!
How would colour depth affect a Alpha render?
Yepp, in particular if your feather exceeds 1024 pixels width. Also naturally 16bit simply provides better precision due to less quantization, which may help improve results in certain situations.
If I remember right you're working mostly with PC's. On the Mac side Prorez 444 is a 10 bit codec with alpha. It's what I use most of the time. If you want the highest bit depth possible then you are stuck with either image sequences, Black Magic, Avid, or some other 3rd party codec. I think that Uncompressed YUV 10 bit is available on Windows machines but I'm not sitting at one right now. Video codecs will give you a reference audio track. Image sequences won't.
Image sequences make it easy to re-render four or five problematic frames and replace them. When you render video this isn't that easy.
If it were me I'd go with at least a 10 bit production codec for everything that needs an alpha and use the same codec to render an alpha only if the codec did not support an alpha.
If you choose to export image sequences, which many production companies work with exclusively you can also add a wav render to the output module which will give you a reference audio track for your image sequence and will not add any noticeable render time to the project. That's probably the track I'd follow for your shot on miniatures, lots of animation, lot's of image manipulation workflow, but I'd have to attend some production meetings first, look at some story boards, and check everyone's experience level before committing to that workflow.