Blackmagic Forum

Here's the intended work flow:

I have a wide UMP 4.6k shot of a person that is ever so slightly soft. Using Super Scale I'd like to scale it up 4X and then reduce if hoping to gain a little sharpness. I tried it and it seems to work but the shot is scaled down in the viewers and the effect, if there is one, is very subtle. And there's no indication except in attributes that anything is happening. It would be helpful to get an idea how Super Scale works to determine when and how it can be useful.

Om 15b3

Well, technically it should, if your machine has the power, but aesthetically?
I did some tests to find out how useful it actually is.

One was upscaling good HD to UHD and compare it to straight UHD from the same camera and scene, both with Super Scale and the standard scaling in DR set to "sharper". This is what I found:

– That camera is doing very well regarding aliasing, but with very fine structures (in my case on fern leaves) both Super Scale and standard produce some moiré, where I don't see any in the UHD footage.
– Clear, angled or curved lines against a contrasting background (in my case palm leaves against the sky) are just a tiny bit sharper and cleaner.

My second test was SD footage scaled up to UHD (!). There seems to be a reason why BM is mentioning exactly that case.
Well, if that footage is interlaced or already has edge sharpening (normally it's even both) just forget it. It'll look ugly. Period. But if you are lucky enough that someone shot progressive and turned down sharpening (or it was a film transfer) it'll look better than the standard process. Somewhat cleaner and more defined, but ever so slightly a mild cartoonish look. This reminds me of the stills software Photo Zoom Pro, but of course I don't know if a similar algorithm is behind this. BUT: this takes about 20 times as much for rendering vs DR's standard scaling.

So, IMHO Super Scale is of limited use, if you have that rare archival material where the conditions above are fulfilled and the extra time can be charged.

What I didn't test is some original footage in 'small' HD (1280 x 720), which is always progressive. It might work pretty well for giving that upscaled version a little edge without oversharpening,

But I'm not sure if you'll gain anything from upscaling and downscaling again other than massive rendering times. After all, without real AI no software can invent details that were not captured in the first place. It might rather 'invent' artefacts, like in my test with the fern.

Thanks, Uli. Much appreciated the info.