Best export settings for long jobs (interrupting/resuming, joining files)

which export settings would you guys suggest for use with jobs that need several days of work to do? i decided to ask in case i’m missing something.
image sequences seem like an easy go-to choice, except the selection of formats and lack of configurability is pretty awful, not to mention that after the job is done, there’s still encoding to be done.
so maybe one of the video formats? i tried exporting as h.265 mkv and using mkvtoolnix to merge files, but for some reason the resulting files have like full one second freezes where the parts start/end, and i have no idea why.
do you guys have this figured out better than i do? if so, how do you do it?

I use PNG because I think TVAI doesn’t have good enough final output options. I need to run another program to make the final video anyway. TVAI converts the colorspace to RGB48le for processing. That’s what PNG uses, so there’s no loss there. I just have to take extra care when making the final video to make sure the colorspace gets converted to something more mainstream.

All that being said, I have never had a job fail halfway through and been able to start where it left off. I’ve had to process the whole movie again every time. I don’t output to higher than FHD. So even a three hour video only takes maybe seven hours at the most—while using the slowest model.

In the past, I have joined folders of images using a batch rename program to make sure they have no gaps. I think the gaps you are seeing are a by-product of how mkvtoolnix joins videos. If I remember correctly, it uses key frames as law. If you have audio, you can often run into the situation where your audio is a bit longer than than the video… but I seem to remember getting those gaps even without any audio.

thanks for your input

png is pretty difficult to use for me, since i’m often upscaling VR content to at least 6K res, which results in tons of space used - i just finished doing a 35 min 60 fps video, and it produced around 1,7 TB worth of images, which is ludicrous. if i ever want to work on a longer video, i’d either have to get like a 4 TB drive/another 2 TB drive, or perhaps compromise quality somewhere along the way. and the most annoying thing is that ffmpeg does support better lossless image formats that could be fast enough to use, like WebP (or maybe even JPEG XL could be tweaked to be fast enough?), but for whatever reason TVAI doesn’t expose them… and yeah, encoding at 6k in timely manner, with decent settings without resorting to NVENC is difficult, too.

maybe you’re onto something with the keyframes, might be worth experiment with perhaps…

If you look around the forums, you can find examples of how to change the encoders.json file to add or modify the outputs you want. It might help.
It would also take forever to encode, but AV1 seems to be the best option for anything 4k and up. I have no idea if VR hardware has embraced that eventuality yet though.

oh i wasn’t aware we can tinker with this ourselves! i will definitely look into it then.

i am aware of AV1 but i don’t really have the hardware to smoothly play such videos, so i am sticking to hevc for now.

1 Like

Hello!
Please write your PC system

CPU
RAM
GPU
OS

sorry, but for what purpose?
if you’re wondering about av1, i made a little oopsie. my gpu can decode av1, it just can’t encode it (got rtx 3000 series). and knowing how god awful slow software encoding is, i’m not leaving my pc on for a full week so i can save maybe like 30% of storage on the output file.

1 Like

my choice - CPU 264/265 and Nvidia 264/265 and deinterlacing - then I use the integrated 770 QSVEnc 264 - VPP- TFF / BFF

I am currently using 4k60fps (maximum)

1 Like

and I use AVC / HEVC 8-bit at 99.99% and 10-bit at 00.01%

Thats really not true that AV1 takes ages to encode.
I can encode at 30 fps - 75 fps using svt-av1 @ crf 18 preset 6.
Thats using av1an on a 10 bit FFV1 file.
That’s on a 5950X.

I use FFV1 10 bit out on TVAI with a mov container.
Colour grade etc etc in Resolve stiil using FFV1 10 bit.
Then as final encode I use svt-av1 for my own use or H265if needed.

i bet a lot of codecs don’t take ages to encode with a 16 core CPU on your disposal lmao
even then, comparatively to other codecs, it does take ages to encode, no way to sugarcoat it

I’m not going to sugarcoat anything.
Just some facts for you.

Source file is 1280 x 720 30 FPS and is 06:24 in length.
Both encoded crf 18 preset 6.

x265: 16 cores

svt-av1: 16 cores

Lets compare lmao 16 cores to 8 cores.

x265: 8 cores

svt-av1: 8 cores

As you can see I don’t consider 10 to 15 seconds longer “Ages”

last time i tested it (several months ago) av1 was at least like 3-4 times slower than hevc, tested with handbrake. did something change so much with this codec in this time?

preset 6 is the fastest preset and with the worst quality, something like x265 preset faster . Let’s see with preset 0 :wink:

You clearly don’t understand or have read what I said :wink:

1 Like

okay so i looked into the whole encoders.json business. i looked into that file just to find a message that this file has been deprecated, so i looked into video-encoders.json instead
i tried to add lossless webp output, so i added this at the end of the file:

  {
    "id": "webp-lossless",
    "encoder": "WebP (lossless)",
    "ffmpegOpts": "-c:v libwebp -lossless 1 -compression_level 4 -pix_fmt rgb24",
    "profile": "8-bit",
    "ext": [
      "webp"
    ],
    "isImage": true
  }

but tvai is having none of that, processing error. when i look into the logs, i see the apparent cause: Unknown encoder 'libwebp'
i guess it’s because tvai ships with a castrated version of ffmpeg? would i be fine replacing it myself with a fully equipped version of ffmpeg, or will something break by doing that?
(pinging @Carlito since you seem to have played around with this before :slight_smile: )

1 Like

TVAI ffmpeg version is compiled exclusively for TVAI so using another version will fail.
To know which video, audio and image formats are available in TVAI ffmpeg do this… Go to C:\Program Files\Topaz Labs LLC\Topaz Video AI
Create a new file and name it Ffmpeg.Encoders.bat
Open that file and add this lines…
@cd /d “%~dp0”
ffmpeg.exe -encoders
pause
Save the changes and exit…
Now run the bat command and it will show you the true…
I’ve already checked and libwebp isn’t available.
Why not use Ffmpeg Ffv1 ? It’s a lossless video encoder codec. Some TVAI users like myself use it.

3 Likes

well this sucks another L for topaz :frowning:
i did hear about ffv1, i can check it out, but i am still looking for an output format as described in the first post - do you have experience joining/merging ffv1 files? it seems like it can be saved as mkv, so maybe mkvtoolnix may also work?

edit: i played with ffv1 and i think i like it. the filesize seems substantially less than png frames while still being lossless. i also tested merging files and it merges them seamlessly. thank you for the recommendation :slight_smile:

edit 2: although one issue i realized with video merging is that it requires twice the storage - which wasn’t really an issue with lossy codecs, but from my initial calculations i could see this being an issue with ffv1. so since my working drive is 2 tb, i can’t be merging files that are larger than 1 tb in total.
i wonder if there’s a clever solution to this, so i can merge larger files. anyone?

1 Like

What is your final goal ?
Are you going to further work on the file in another program ?