https://bugs.kde.org/show_bug.cgi?id=498263
--- Comment #7 from [email protected] --- (In reply to lras916 from comment #6) > I would like to say that I am also encountering this issue. Even when trying > to export lossless video using one of the "lossless/HQ" render options, the > output is still squished down to 4:2:2 with the "ffv1" and "utvideo" > options. I even tried to create my own render preset with utvideo and > specifying "pix_fmt=rgba" as an additional parameter. The rendered video is > in "rgba" format, but the image looks like 4:2:2. This makes me think that > something somewhere else down the chain converts the images to 4:2:2. > > A finding I made was that the "Video with Alpha">"FFmpeg with FFV1" actually > DOES render losslessly. And I think it's all because of the parameter > "mlt_image_format=rgba". I then tried adding this parameter to a modified > version of the "Lossless/HQ">"Ut Video" preset and suddenly that renders > losslessly as well. > > I definitely agree that this is a bug that should be fixed, but as a > workaround, OP should add "mlt_image_format=rgba" to their list of > parameters. Thank you for checking this out. I can confirm that adding "mlt_image_format=rgba" to my custom preset (with pix_fmt=yuv444p in my case) fixed the issue. I did some more research and found this comment by the creator of MLT regarding their stance on the issue: https://github.com/mltframework/mlt/issues/398#issuecomment-452977479 As the default will not be changed upstream, I think it would be a good idea to make sure all the rendering presets under "Lossless/HQ" contain mlt_image_format=rgba, like all the presets under "Video with Alpha" have. When 10-bit support is added in the future, I think it would be nice to have a GUI drop-down menu next to "Render full color range" called "Pixel format" that lets you choose from "yuv420p" to "yuv444p10le", for example. To implement this, you would need to check the encoder's supported pixel formats. You can use ffmpeg -h encoder=(encoder name, ex. libx264) to check this. The default should be whatever the selected preset specifies. This could also be split into 2 options ("Bit depth" and "Chroma subsampling"), but I prefer "Pixel format" since it ties directly to the corresponding FFmpeg option. -- You are receiving this mail because: You are watching all bug changes.
