Hello! I’ve been doing this a lot in the past 3 months, making tests before I embark on a mass conversion.
I did actually try to get started with a scheme that did HDR at the end, so that’s: SD->HD, HD->AI (4K), SDR->HDR
But then I started feeding low-resolution HDR footage (just 1080p or 576p) directly into Topaz, and the results were magnificent. The colours and the details just snapped more realistically. (EDIT: One surprise — the preview UI on Topaz app doesn’t show any HDR at all. But the outputs work 100%! So just have faith, or if you have to live and die by the evidence, then only judge the output file appearance from an app that knows how to show HDR — and sadly that’s not including VLC)
So: I’d refer you to the golden rule of GIGO and BIBO (Brilliance In, Brilliance Out) — give Topaz the best possible quality, and it will do wonders. The only caveat is that’s it’s a bit better if you match the AI model to the resolution it’s “native” to, and you can look that up in the detailed model descriptions. For instance, Iris v1 works really well with 480p sources.
Now to your case — your sources are 1080p and if they’re “truly” 1080p and capture plenty of detail, i.e. professional quality footage, all in focus etc, then by no means downscale — it would delete details. And anyway you’d be using a much more suitable AI model than Iris v1. I’ve used Gaia HQ on some very high-quality 720x576 material after I moved it to 1920x1080 and it looks magnificent, but all the source material was studio cameras and really top-end material. Even if I didn’t use AI, there was plenty of extra detail to “suck out” of the source format, before presenting it online (and compressing it to smitherines for a streaming platform).
So: Create your intermediates wisely.
- Tidy up your 1080p SDR footage if it needs it — any cropping, cleaning, hopefully none
For your 720p footage, try scaling it up with custom sharpening. Do an end-to-end test.
Then run another test with keeping it 720p until step 3 where you feed it into Topaz.
Eyeball the results and decide whether to unify everything at 1080p as an intermediate.
I suspect you can make a 1080p non-AI “master” … non-“RAW” cameras share the same look.
Step 4 is Step 1a.
4. Unfortunately I doubt your SD-480 footage (4:3?) can participate like this.
You’ll probably need to separately process these after you’ve created an HDR-4K Topaz output.
It depends how much you want to care about hiding the fact it’s SD-480 source?
Your workflow either way is going to need you to make a 1080p SDR and 1080p HDR intermediate.
Throw all your SD-480 material in there as if you didn’t have Topaz, judge the output on its merits.
Then at the end, with however many short clips you have of SD-480, put the originals together;
create a 480p HDR intermediate, then throw it into Topaz to get 4K HDR.
NOTE You will need a different AI model. Make sure you choose a better one for MQ or LQ.
Insert that video into your 4K HDR master and see if it matters enough.
If it doesn’t, then forget about Step 4 / Step 1a.
-
Upmap it to HDR using custom LUTs, e.g. Ben Turley’s LUTCalc.
Make an “intermediate” product you’re happy with.
And it really helps to use a 1500+ nits HDR screen.
You think you know what you’re getting with 600 nits previews, but you don’t.
-
Once you have the “best” 1080p HDR material, feed that into Topaz and ask for 4K.
Try a few models, focussing on the HQ ones.
Now watch the results on your 1500+ nits HDR screen. You’ll be stunned.