Best FFmpeg Presets for HEVC Encoding With AV1 Fallback on Mid-Range GPUs

[crush]

Hi everyone,

I’ve been using FFmpeg for a while to encode video, and I’m now trying to dial in a workflow that balances quality, speed, and compatibility. My goal is:


Encode most of my library in HEVC (x265) with sensible file sizes.
Provide an AV1 fallback option for devices that support it.
Keep the process fast on a mid-range GPU (RTX 3060 level).


I’m particularly interested in recommendations for:


Preset and tune combinations that give a good balance of quality and encode time for HEVC using NVIDIA NVENC.
Equivalent setups for AV1 (SVT-AV1 / rav1e / GPU) — what codecs and settings are most practical right now?
Real-world examples of command lines you use — especially where you generate two outputs from the same source (HEVC + AV1) with minimal re-processing.


I’m aware of the bitrate ladder approach and CRF targets, but I’d love concrete examples and reasoning behind suggested settings. For instance, how you choose QP/CRF/bitrate targets for HEVC compared to AV1 for similar perceptual quality.

Thanks in advance — keen to see what the community uses as a workflow here.

Emily

[Z2697]

HEVC is the codec, hevc_nvenc and x265 are encoder implementations of the codec.
Which one exactly are you asking?

[GeoffreyA]

For AV1, try the SVT-AV1-HDR fork (https://github.com/QuickFatHedgehog/FFmpeg-Builds-SVT-AV1-HDR/releases), which has good quality and handles grain tolerably well. The classic AV1 encoder, aom, used to have the best quality, but is less important today and terribly slow.

-c:v libsvtav1 -preset 4 -crf 30 -svtav1-params tune=0

x265 is the most popular HEVC encoder; it needs a bit of tuning to get safe quality. Start off with:

-c:v libx265 -preset slow -crf 18 -x265-params no-sao=1:deblock=-1,-1:aq-mode=1

Tune from there. Depending on whether the material is 1080p, HDR, or grainy, different approaches can be taken. If the file is too big, try CRF 21 and see what it looks like. When comparing settings, use two passes at the same bitrate. For example:

ffmpeg -i SRC.mp4 -c:v libx265 -preset slow -b:v 3000k -x265-params pass=1 -f null -
ffmpeg -i SRC.mp4 -c:v libx265 -preset slow -b:v 3000k -x265-params pass=2 TEST.mp4

x265 and SVT-AV1 give better quality per bit than hardware encoders, but today's GPUs do a respectable job and can save time. Perhaps the biggest challenge is encoding grain, where denoising is often needed. Unfortunately, this is not a solved problem.

[benwaggoner]

Hi everyone,

I’ve been using FFmpeg for a while to encode video, and I’m now trying to dial in a workflow that balances quality, speed, and compatibility. My goal is:


Encode most of my library in HEVC (x265) with sensible file sizes.
Provide an AV1 fallback option for devices that support it.
Keep the process fast on a mid-range GPU (RTX 3060 level).


Those are a hard set to do well at the same time!

What are you source formats and bitrates? Unless they are almost lossless or something ancient like DV25, reencoding enough to save significant bits almost always will have at least a minor visible quality loss. Actually saving bits while preserving source quality making both a HEVC and a AV1 is somewhere between "impossible" and "very slow and hard." Doing it with GPU encoding makes the very slow part better, but also loses some compression efficiency compared to a good software encode.

So, big picture, what are you trying to accomplish?

And is it something that buying a >10 TB hard drive for extra storage would accomplish a lot easier and cheaply? Just the electricity costs for doing this (and then the AC for getting rid of the waste heat of doing it when it gets hot, and this could easily last for seasons if you have a lot of content and you're trying to do an optimal job of it) might be more than the cost of a drive depending on where you live.