This is a discussion of a technique for getting optimum detail rescaling using Proteus. Your participation, and suggestions are welcome.
For my experience Proteus is able to recover some details by adding a tiny bit of de-halo, because the de-halo is based on an older Artemies mode that was more capable of recovering details than the later ones. But don’t ramp the de-halo up by too much or your video gets a dreamy shiney look. I personally add up to +20 de-halo in Proteus.
Hi, can you demonstrate ‘recover details’ by posting comparison images? I like dehalo for getting rid of over-sharpened artifacts, but I see that it softens surface details, rather than enhancing them. Would like to see your examples.
I will try as soon as possible. Proteus is not generating previews for me while being stuck at “Model Loaded” 0s in this version.
Did you get 22.214.171.124b. And do you have a 30 series card. If so did you get the 30 series install? If not did you accidentally get the 30 series install for a non 30 series card
at first i had the wrong version for my RTX 20 series, but now its the win all version. I dunno why it doesn t work.
It’s all very electric and dangerous…
I was hoping to get some advice on workflow to bring DSLR safari footage 720p or 1080p up to 4k. I like proteus relative to auto to boost details and dehalo, adding grain helps as well but I do all this in one go. Also adding low value stabilisation with reduce jittery motions AND Rolling shutter Correction tends to find more details due to interframe detection. What you guys use?
Obviously this depends on the source footage but I was having an issue where the source (1080p) had a lot of noise/grain and proteus was destroying details on faces (wrinkles, freckles etc.) to remove it. I found the best way was to first do a 1x pass of Artemis HQ to get rid of a lot of the noise but preserve the details, and then run that through the 2x Proteus with some fine tuned manual settings. This resulted in Proteus being able to remove the rest of the noise and enhance the detail instead of removing it.
Nice topic. Please share your workflow
I made the comparison today. Please have a look onto the green parts, especially in the lower green of the video that is being slightly more detailed in strong dehalo (often medium halo is strong enough):
keep original size:
upscale 2 x
upscale to 4k:
Of course it depends on the scene again. In the 4k upscale the dehalo did not a great job by blurry out details, but in the first one it was slightly better than the others.
Thank you for posting this. It’s a good demonstration and I see what you mean. My material is more people focused rather than landscapes and architecture, so in my case, I find none of the Artemis models retain the detail I would like, dehalo included. Things like hair, clothing, skin.
I do use the older Artemis 6, 7, or 8 models sometimes (in VEAI 2.x) and those do a better job on some types of footage, but they don’t clean as well.
Do all the sliders of Theia work for you?
yes, I can move them all.
Okay but do they do anything? Only Reduce Noise applies sharpening for me. The others don’t do anything.
Not trying to hijack from Proteus, just I’ve been waiting for Theia to work again so I can see how well it works in combination with Proteus.
You are right, Theia in general feels quite conservativ in terms of improving the video footage. As far as I understood the Theia was originally designed to recover details by looking into several images of the video, but the results are often not very impressive.
I dont understand why is Theia still kept in Video AI, I havent found a single case in which Theia did anything to any of my videos
Stay away from Theia, and your problem is gone (btw no problem with Theia on my PC)