Decisions for Commercial Project

Topaz Video General
1

I may have an opportunity to enhance a collection of B&W films from the 1930’s and 40’s. Not sure yet what ‘original’ file type will be provided but probably ProRes 1080p and in some cases, 2K. These films when scanned from a variety of sources which were good to fair in terms of noise, focus, and grain.

I have determined that some files will only need an Artemis or Proteus pass, but others will require a Starlight model.

They are requesting these seven average Size/Bitrate (constant) outputs in the final delivery:

416x234 - 363K
416x234 - 89K
640x360 - 1241K
768x432 - 2107K
960x540 - 2843K
1280x720 - 5637K
1920x1080 - 8966K

My question revolves around the Starlight processing. I’m thinking I should I downsize the 1080 to something smaller before running. Is my thinking correct when considering processing time? (I’ve got two 5090’s to work with.) Really don’t know where to start.

Any help appreciated.

2

Reading into this, they probably are aiming for an adaptive bitrate ladder for use with HLS or DASH.

Focus on producing one, high quality mezzanine file - to the best quality that you can. That becomes your new Golden Original. 1080 / BT.709 etc. Perfect gamma curves. Talk to them about what codec want the new master in (FFV1, ProRes, MJPEG or and if you wanna go real-pro, DCP or MXF container with appropriate codec). That should be your deliverable - and should be “archive” quality - and if it is B&W you only need the luma component (the chroma component is just noise). Spend all your time, love and processing power on making this the best quality, adhering to studio and broadcast standards. Capture the full-frame, because it is unlikely that these were even 16x9. Forget about the size and the storage. This is what you should be paid for.

As for the multi-bitrate “final delivery” specs… don’t waste Topaz on these. Just use FFmpeg, Apple Compressor, AWS Media Encoder, Resolve, Adobe. These should be facets of the main master/mezz deliverable, and considered temporary… So that when the next consumer-grade codec comes along you can create from your master/mezz.

To a broadcaster, studio or streamer, these multi-bitrate variants are disposable - they can be generated in near-realtime with any-old-off-the-shelf H.264 encoder. There will be things like pixel formats, key-frame alignment and GOP settings that will need repeated tweaking to meet the details of the streaming spec. You cannot be doing yet another Topaz pass every time there needs to be a tweak.

This is pretty standard practice. You could even go as far as producing a master (archive quality, lossless FFV1) and a mezzanaine (lossy, but good enough)

3

Thank you for the detailed response. You have provided a lot to chew on and obviously I have to educate myself on the spec’s you mentioned.

These films were obviously shot on 35MM with an aspect ratio of 1.375:1, although some may be 1.33:1. I’m hoping that’s what they will offer me. The sample resolutions I mentioned in my post all have black bars left and right as they are intended for streaming.

For Starlight, should I downsize my master to 540 first?

4

I’ll let the others chime in with whether Starlight would benefit from downscale, but I would typically get hold of the best quality original that I could lay my hands on, feed that into Topaz and produce the best quality, square-pixel, BT.709, 1080-height output that you can. Width will vary based on aspect ratio (CinemaScope etc).

Put all your effort and processing power into getting that golden (re)master looking right - no black bars, perfect aspect ratio, crisp upscale, respect the film grain (!) etc. Primary goal is to produce that golden, high quality version - burn your GPU cycles on producing the best version that you can with the best model that respects the technicals, as well as creative aesthetic (such as grain, which is an important part of the aesthetic). You may find that some of the models look too-smooth, too-sharp grain. Respect the “period correct” creative aesthetic - the aesthetic of a 1940s B&W movie is very different from what you would want Topaz to do to modern animation or gameplay. You may want to consider minor dust & scratch removal with something like Neat Vdeo - but a little bit of dust & scratch is also part of the aesthetic.

Then, when it comes to the multi-resolution facets, you can apply 16x9 crops, pads as per specs with off-the shelf distribution transcodes.

Sounds a exciting project - and one where you will have to walk the line between preservation and restoration.

5

More great tips - I appreciate your expertise. I’m just an amateur who enjoys tweaking films for my own enjoyment. ‘Pleasing’ output up to my standards has been the rule…until now.

The single overarching concern I have is having to run the original file through multiple models and apps to achieve that new, “Golden Original”. How do I keep the file as lossless as possible?

I think I understand the obvious guidelines: use ProRes or DNxHR output, don’t upscale, keep bitrate constant, maintain the same frame rate and resolution, etc. And if there is any ‘cleaning’ to be done (I have DaVinci and Neat Video), leave Topaz for the last pass(es).

Anything else you can add?

I do believe that the highest quality can be achieved by bypassing the GPU and relying on the CPU, at the great cost of time. Not something I have ever tried. Any experience at this level of restraint?

6

Disc space, disc space, disc space is gonna be your thing if you really want lossless.

Personally, I like FFV1 (level 3) codec as my intermediary codec. It is lossless, but not tooo heavy. You should be able to do any of your preprocess in Resolve, save as FFV1. Process in Topaz, again save as FFV1. [ffv1_vulkan GPU hw shader is an interesting recent development]

FFV1 is a bit of an oddball codec and not supported everywhere, but it is loved by archivists and a ref of us restoration amateurs. Others like one of the higher-quality versions of ProRes, which are lossy, but accelerated on Apple Devices. Visually, there’s not much in it.

Resolve > FFV1 > Topaz > FFV1 > FFmpeg
Resolve > ProRes > Topaz > ProRes > FFmpeg

I like FFV1 because it supports gray10le or gray16le pixel formats which really help on the B&W stuff - you only need the Y component of the YUV video. It all gets a bit FFmpegy and nerdy, but in many tools you can use grayscale pixel formats, or “filter/extract planes” to focus just on the luma / greyscale. It gets expanded back up to color when you go through Topaz’s RGB, but when you convert back down to YUV, you can once again dump the chroma - reducing the required space and usually increasing speed - and removing color-casts. As you can tell, I’m a bit of a shill/stan for FFV1, but it is not for everyone.

I would avoid any hardware accelerated versions of H.264 or H.265… They are typically optimized for low-latency rather than quality. As you say, for H.264 and H.265, you want to be using x264 or x265 software encoders. Worry about that for your output and distribution formats.

Your summary is reasonable - other than the “use constant bitrate”. You should be targeting a quality (or even lossless). Bitrate should never be a constraint for your high-quality internal workflow, unless you are short on storage.

As you say, so all your prep-work in the NLE. Do your crops and your aspect ratios, clean up your dust and scratches. Then throw into Topaz to get your best upscaled version. And something like resolve+x264 or ffmpeg+x264 for your final deliverables.

7

I would also say run 4:4:4 when possible for intermediates. When doing rescales during projects it does matter for the final sharpness and quality…

8

For storage, I’m OK at 100 TB plus. Picked up two 22 TB Segate’s back in February for $255. each. I knew what was coming…

Thank you for the additional insights. Lot’s to consider and test.

9

Another great suggestion. :grinning_face: