Posted by qbow883 12/27/2025
These days I'm much more inclined to try and transparently encode the source material, tag it appropriately in the media container, and let the player adjust the image on the fly. Though I admit, I still spend hours playing around with Vapoursynth filter settings and AV1 parameters to try and get a good quality/compression ratio.
I have to say that the biggest improvement to the experience of watching my videos was when I got an OLED TV. Even some garbage VHS rip can look interesting when the night sky has been adjusted to true black.
Given the increasing abilities of TVs and processing abilities and feature sets of players, I'm not much persuaded to upgrade my DVD collection to Blu-Ray. Though I admit some of that is that I enjoy the challenge of getting a good video file out of my DVDs.
I partially disagree with the use of ASS subtitles. For a lot of traditional movies, using SRT files is sensible because more players support it, and because it's often sensible to give the player the option of how to render the text (because the viewing environment informs what is e.g. the appropriate font size).
My impression is, their audience equates file size with quality so the bigger the file the more "value" they got from the creator. This is frustrating because bigger files means hitting transfer limits, slower to download, slower to copy, taking more space, etc...
Unless one lives in a country where the internet is slow and/or hard drives are expensive, I think the audience does not care.
My hypothesis is that they use a really high quality value, and that there are diminishing returns there.
ffmpeg seems ridiculously complicated, but infact its amazing the amount of work that happens under the hood when you do
ffmpeg -i input.mp4 output.webm
Edit: from some googling it looks like encoding is encoding, whether it’s used for recording or rendering footage. In that case the same quality arguments the article is making should apply for recording too. I only did a cursory search though and have not had a chance to test so if anyone knows better feel free to respond
GPU acceleration could be used to accelerate a CPU encode in a quality-neutral way, but NVENC and the various other HW accelerators available to end users are designed for realtime encoding for broadcast or for immediate storage (for example, to an SD card).
For distribution, you can either distribute the original source (if bandwidth and space are no concern), or you can ideally encode in a modern, efficient codec like x265 or AV1. AV1 might be particularly useful if you have a noisy source, since denoising and classification of the noise is part of the algorithm. The reference software encoders are considered the best quality, but often the slowest, options.
GPU is best if you need to temporarily transcode (for Plex), or you want to make a working copy for temporary distribution before a final encode.
I think it might be one of those classic “everyone should just get good like me” style opinions you find polluting some subject matter communities.
The vlc was how you could get any movie to work (instead of messing with all these codecs, which apparently, in lieu to another comment in this thread, aren't really codecs).
Grump grump grumpity grump. Same experience with every dashcam I've bought over the years.
The idea that YCbCr is only here because of "legacy reasons", and that we only we discard half of chrominance because of equally "legacy reasons" is bonkers, though.
Similarly, chroma subsampling is motivated by psychovisual aspects, but I truly believe that enforcing it on a format level is just no longer necessary. Modern video encoders are much better at encoding low-frequency content at high resolutions than they used to be, so keeping chroma at full resolution with a lower bitrate would get you very similar quality but give much more freedom to the encoder (not to mention getting rid of all the headaches regarding chroma location and having to up- and downscale chroma whenever needing to process something in RGB).
Regarding the tone of the article, I address that in my top-level comment here.
It really isn't. You have to scroll 75% of the way through the document before you it tells you what to actually type in. Everything before (9000+ words) is just ranty exposition that might be relevant, but is hardly "quick".