What is the general consensus on multiple upscales?

There’s a 480p video that I would love to upscale to 4k, even with all of the artifacts. What I’m wondering is, might it work better if I go from 480-720-1080-4k, rather than just 480-4k? theoretically in my head it seems like it would have a more accurate effect in stages.

Has anyone tested this to see?

Thank you :slight_smile:

makes sense to me, I’ve been wondering the same thing and came here to see if anyone else knew.

I have a 320x272 avi clip that was taken on a nameless phone 15 years ago.
Using the Proteus model and auto detecting new setting parameters each pass, I went up to 1920x1080 (No crop) by increasing 200% each time. Then I did one straight to 1080 to compare.

Ummm… they both had similar looking artifacts but not in the exact same places.

Maybe I should have tested on a better quality clip.

I found another clip I made on Blender. Similar resolution, but much better quality.
Multi pass made it seem more oil paint-like.
One shot was more blurry around the edges, and less oil paint-like.

I think a test of a more common movie still needs to be done.

If it were me – any pre-processing work on the 480p clip should be done first to give VEAI a good original – then upscale to 720p. Then you might need to correct issues with the new clip at 720p before upscaling to 4k or whatever. At 720p, you have a high-res version, probably with a lot of artificial filler. I would think that the software can only do so much and eventually you hit a point of diminishing returns.

Results will of course vary significantly based on your source material, and what types of issues it has, but I’ve recently been working with 1080p footage that has a lot of compression artifacts because it was captured from a livestream (presumably hardware encoded, or with a fast preset to enable realtime streaming), and have been trying to upscale it to 4K.

In my scenario, and using Proteus Fine Tune V3, I’ve found that doing two-passes is giving me better quality. In the first pass I just apply compression reversion and a bit or deblur. Then in the second pass I apply noise reduction. I found that when I was doing it all in one step, noise reduction would remove a lot of details and leave an artifically smooth look on a lot of things, but if I let Topaz upscale first and then do noise reduction in a second pass, it significantly cleans up my clip, but without removing lots of details.

I’m going to have to try that approach. I have a few clips that I’ve just given up and settled for some of the aspects looking good, but others, not so great. Given up, because if I tune it to make the others look good, then I lose the first aspects looking good.

If you’re really serious about getting the best possible result, then you could mitigate that issue by splitting up your clip. I haven’t yet gone that far with mine, but I’m considering it, because I really want to turn up sharpening, deblur and detail recovery on the close-up shots, but turn them down on zoomed out shots where those parameters introduce some horrendous looking artifacts.

Also if you’re doing multiple upscales with Topaz then you should ideally output as images because the approach that Topaz uses to encode your video results in quite poor quality compared to doing it yourself. If you’re going to feed the output back into Topaz then you end up unecessarily losing quality if you let Topaz encode after the first pass.

Just doing a simple -preset slower with ffmpeg I found that I was getting an mp4 half the size of what Topaz gave me, and in better quality. Fortunately this is something that the beta is now giving us control over, but for the current release version you are stuck with Topaz’s default which isn’t great.

I’ve been putting off doing it that way, but yes, it may come to that.
Humm, I have Topaz set to CRF 17 mp4, and not noticed any loss yet, but have only been doing one pass. I worry that I don’t have the drive space for images right now. Either way, ffmpeg is needed to make the final result a reasonable size.

From my experience, the best results are from doing X2 or X4 upscale. So, try to go with Dione 480p to 960p, then go to 4K with Artemis or something.

I agree. This approach is giving me great results. Better than one pass, for sure. Sadly, it takes about three times the amount of processing time.