Lossless Encoding Comparison 2024

[tormento]

Time for another comparison:

This commit (https://git.ffmpeg.org/gitweb/ffmpeg.git/commitdiff/ed2391d3410e253bf37e5e7274cc945c866bd337) implements a standard, compliant, version 3 and version 4 FFv1 encoder, entirely in Vulkan. The encoder is written in standard GLSL and requires a Vulkan 1.3 supporting GPU with the BDA extension.

[Z2697]

--output-depth is a CLI only option. To "control" the internal bitdepth used in FFmpeg libx265 you need to convert it. I suggest using a combination of zscale and format in filter (e.g. -vf zscale,format=yuv420p10le) because the default swscale conversion may do something weird. The format filter after zscale only sets output format for zscale filter and does nothing else, the whole conversion is done by zscale.

Same thing with FFmpeg libx264, actually. This is not a bug. It's by design.

--input-depth tells the reader the bitdepth of the input YUV file because it has no metadata, the reader will read anything according to what you told it.

The actual "pad/truncate the bits" stuff happens when --output-depth (it also sets internal depth) is different from input depth, specified by hand or determined by reader (Y4M/AVS/VPY/etc.).