What bitrate and format for editing old video footage

Hi 8 footage is lower quality than DV . . . only marginally better than VHS so I have found no point in using anything higher than encoding at 480p in H.264 at 2500kbps.

Do a short (1 minute) test using that setting and then try the highest settings available.

Examine identical frames, highly enlarged in FCP X and see if you can spot any difference.

I like ProRes LT. It would be good. For small file size HEVC 10000 kbps should be good.

I would think those numbers are too high for HEVC, which a really efficient codec, hence the E. ProRes is really easy for FCP to work with because it’s a I-frame codec. HEVC is more processor intensive and if the work gets complicated, layers, effects, graphics, PRLT is easier.

> so an hour of DV would be around 30 hours

Ouch.

A few years ago I encoded my (archived from iMovie 1.0.2-6.0.3) 25 fps interlaced PAL rectangular pixel 720x576 .dv files to bob-deinterlaced 50 fps progressive H.265 square pixel 768x576 .mp4 with ffmpeg (installed via MacPorts). IMHO the quality is very good and I could not get as good quality output from Compressor although I did not thoroughly test all its options.

That ffmpeg process in macOS 13.3.1 Ventura now takes about 3 minutes per 1 minute of .dv with the settings below.

I did some tests with my poorest quality input (i.e. poorly-lighted moving detailed input) and decided to use -crf 18 -preset slow and let it run all day and night on my Mac mini 2018 in ffmpeg .dv-to-.mp4 batches (let me know if you need corresponding batch settings).

ffmpeg defaults are '-crf 28 -preset medium'. Setting them up to '-crf 51 -preset ultrafast' (8 seconds for 1 minute of .dv) and skipping all that deinterlacing and scaling is MUCH faster but I preferred to do this with the best quality without spending unnecessarily time.

ffmpeg -i input.dv -vf bwdif=1,scale=788:576,crop=768:576:10:0,setsar=sar=1/1 -c:v libx265 -crf 18 -preset slow -timecode 00:00:00:00 -tag:v hvc1 -c:a aac -b:a 128k output.mp4

I am a firm believer in testing things for myself and have frequently found that doing everything by the book for maximum theoretical quality does not necessarily produce that quality.

Forget the theory and do a test for yourself.

It won't take long.

I have converted DV footage at 13GB per hour to H.264 540p at 2500kbps which gives files around 10% of the size and under extreme magnification in FCP X have never been able to spot any difference.

Of course, I could be wrong but you can test it for yourself. Maybe you will spot a difference or maybe not.

There is no point in converting to a higher resolution . . . it will just produce much larger files with little, if any improvement. Anyway, test a sample to see what works for you.

That is similar to the AI that I achieved with HitPaw.

There is a big temptation to upscale the video but in my brief tests there was no noticeable improvement over using the original resolution plus the downside was that the encoding took much longer and the file sizes were considerably larger.

So test an original resolution with an upscaled one to see if there is any tangible advantage.

I was sceptic about upscaling and bob-deinterlacing rectangular pixel .dv footage to some odd square pixel non-integer resolution (i.e. PAL 720x576 to some other vertical square pixel resolution other than 768x576 or 788x576 or vertical resolution 576, 1152 etc) with double frame rate.

But that looks very good. Unless there are some AI-generated artifacts?

But how long does it take to encode 60 minutes of .dv to H.265/H.264 with that on some newish Mac?

I have a 2017 iMac.

On a test I did a few weeks ago my 4m 38s DV video would have taken an estimated 2h 33m so an hour of DV would be around 30 hours which I consider to be ridiculously long.

If the M1/M2 computers are as fast as the benchmarks suggest, the time would be down to 10 hours.

Upscaling to 1080p or 4K would take proportionately longer and in my tests there were no discernible qualitative improvements.

You have to ask yourself whether it is worth the time and effort. Most of us naturally want the highest possible quality but the sad truth is that most of our relatives and friends, like Rhett Butler, "couldn't give a ****".

Are there any rivals for TopazVideoAI?

...

I did TopazVideoAI 3.2.3 tests with my oldish Mac mini 2018 which seems to use some ffmpeg CPU and almost 100% GPU for this:

1 minute PAL 720x576 .dv (216 MB) to bob deinterlaced 50 fps H.265 768x576 .mov (135 MB) 47 minutes.

1 minute PAL 720x576 .dv to bob deinterlaced 50 fps H.265 1920X1080 .mov (625 MB) 68 minutes.

...

My previous 1 minute PAL 720x576 .dv to bob deinterlaced 50 fps H.265 768x576 .mp4 with the slowish '-crf 18 -preset slow' ffmpeg options with that Mac mini 2018 is 3 minutes.

...

AFAIR my old 1 minute PAL 720x576 .dv to 25 fps H.264 768x576 .mp4 with the default MPEG Streamclip "iTunes preset" option with Mac mini late 2009 OS X 10.11 El Capitan with the same .dv input was somewhat worse quality with about 1:3 speed.

...

So maybe I try this after a few years in a new Mac with AI. Especially when the current Topaz $300 is somewhat too pricey for me.

...

p.s. In my work I have used some newish iterative reconstruction algorithms and finally decided to put their settings "to eleven" because despite the warnings that "the output might then be too plastic". IMHO it was better than the medium quality setting with more grain in noisy input.

...

Below are screen captures from a good-light but moving interlaced PAL .dv input to 1. ffmpeg SD, 2. Topaz SD and Topaz HD H.265 output (SD 576 scaled to 1080 with GraphicConverter in these images. Topaz uses slightly different rectangular pixel aspect ratio than I, my setup is to horizontally scale PAL 720 to 788 and then crop to 768).

With this well-lit input Topaz is slightly better than ffmpeg but Topaz HD is not much better than SD when viewed at 100 %.

With grainy input Topaz is clearly better with no introduced artifacts in my quick tests.

What’s the frame resolution and the frame rate?