x265: --output-depth Needs to be set?

[Forteen88]

I wonder, do I need to set:
--output-depth 10

if I set:
--profile main10
? Or does main10 set it automatically?

UPDATE: Thanks FranceBB.

[FranceBB]

There's no need to specify --output-depth 10, Main10 encodes everything in 10bit automatically.
For instance, if you feed x265 with an 8bit stream, it will dither it up to 10bit and encode as 10bit.

[Boulder]

Does it make sense to use --dither if the input clip is 16bit and output 10bit?

[FranceBB]

Originally Posted by Boulder

Does it make sense to use --dither if the input clip is 16bit and output 10bit?

It does.
x265 will accept input pixels of any depth between 8 and 16 bits regardless of the depth of its internal pixels (8, 10 or 12) and it will shift and mask input pixels as required to reach the internal depth.
There have been many people arguing that it's better to let x265 dither the input and other that kept saying that it was better to frameserve the desired output bit-depth.
I think that if you filter out issues in 16bit, it's actually better to frameserve 16bit to x265 'cause this way it's internally aware of the "original" 16bit and it will dither it down in a way that it makes easier to compress while preventing banding.
If you dither it yourself using a frameserver (let's say 8bit source, 16bit filtering, dithered down 10bit output), x265 will think that it's a "normal" 10bit input and it will deal with it as it is and - according to the way you dithered it down - it might require more bitrate.
This is most noticeable with 8bit, but mathematically it's true for 10bit and 12bit as well.
In other words, instead of fine-tuning your dithering method trying to check the differences between different dithering methods available (ordered dithering, Serpentine Floyd-Steinberg error diffusion, Stucki error diffusion, Atkinson error diffusion etc), just let x265 handle the dithering for you: it will do it in the best possible way.

[benwaggoner]

Originally Posted by FranceBB

There have been many people arguing that it's better to let x265 dither the input and other that kept saying that it was better to frameserve the desired output bit-depth.
I think that if you filter out issues in 16bit, it's actually better to frameserve 16bit to x265 'cause this way it's internally aware of the "original" 16bit and it will dither it down in a way that it makes easier to compress while preventing banding.

x265's --dither mode isn't that advanced. It would be cool to have dithering driven by RDO, ala trellis quantization. But that's not in any encoding tool I know of.

If x265 is going down to a lower bit depth, using --dither will improve quality vs. the default dithering/truncation mode. But --dither isn't magic; a well-tuned external ditherer could well outperform it. But for when that isn't available, having --dither available is welcome.

FWIW, in some recent experiments I found that ffmpeg's x_dither does a better job of reducing banding in 10-bit to 8-bit conversions than x265's --dither.