/FFmpeg-CRT-transform

CRT simulation without shaders... the slow way

Primary LanguageBatchfile

FFmpeg CRT Transform

Windows batch script for a configurable simulation of CRT monitors (and some older flat-panel displays too), given an input image/video.
Requires a git-master build of FFmpeg from 2021-01-27 or newer, due to a couple of bugfixes and new features.
See https://github.com/viler-int10h/FFmpeg-CRT-transform/ for the latest version.

Usage and Configuration

Syntax: ffcrt <config_file> <input_file> [output_file]

  • input_file must be a valid image or video. Assumed to be 24-bit RGB (8 bits/channel).

  • If output_file is omitted, the output will be named "(input_file)_(config_file).(input_ext)".

  • How to configure: all settings/parameters are commented in the sample configuration files, which you can find in the "presets" subdir.
    NOTE: the included presets aren't guaranteed to accurately simulate any particular monitor model, but they may give you a good starting point!

Tips

  • Input is expected to have the same resolution (=storage aspect ratio, SAR) of the video mode you are simulating, including overscan if any.

  • The aspect ratio of the simulated screen (=display aspect ratio, DAR) is not set directly, but depends on the SAR and on the pixel aspect ratio (PAR): DAR=SARĂ—PAR. The PAR is set with the PX_ASPECT parameter.

  • The aspect ratio of your final output is set separately with the OASPECT parameter. If it's different from the above, the simulated screen will be scaled and padded as necessary while maintaining its aspect ratio, so you can have e.g. a 4:3 screen centered in a 16:9 video.

  • Processing speed and quality is determined by the PRESCALE_BY setting. This also affects FFmpeg's RAM consumption, so if you get memory allocation errors try a lower factor.

  • Most of the processing chain uses a color depth of 8 bits/component by default. Setting 16BPC_PROCESSING to yes will make all the intermediate steps use 16 instead. That makes the process twice as slow and RAM-hungry, but if your settings are giving you prominent banding artifacts and such, try going 16-bit.

  • By default the output colorspace is 24-bit RGB (8 bits/component), but you can change that by setting OFORMAT to 1: for videos, this will output YUV 4:4:4 at 10 bits/component. For images, you'll get 48-bit RGB (16 bits/component), which works with .png or .tif for instance.
    (Of course, to get the most out of this, you'll want 16bpc processing as mentioned above)

  • In general, speed is the weakest link in this whole thing, so you may want to test your config file on a still .png image (or on a few seconds of video) first, tweak things to your liking, and tackle longer videos only after you've finalized your settings.

Write-ups, videos, sample images

  1. Color CRTs: https://int10h.org/blog/2021/01/simulating-crt-monitors-ffmpeg-pt-1-color/

  2. Monochrome CRTs: https://int10h.org/blog/2021/02/simulating-crt-monitors-ffmpeg-pt-2-monochrome/

  3. Flat-Panel Displays: https://int10h.org/blog/2021/03/simulating-non-crt-monitors-ffmpeg-flat-panels/