Best Practices for Upscaling 1080p to 4K - Software and Techniques?

[NickShaw]

Hi everyone,

I've been experimenting with upscaling some of my older 1080p video content to 4K and I'm looking for advice on the best software and techniques to achieve high-quality results. I've tried a few methods, but I'm not completely satisfied with the outcomes.

What software do you recommend for upscaling video files from 1080p to 4K?
Are there specific settings or filters that you find effective in preserving detail while minimizing noise?
Does anyone have experience with AI-based upscaling tools, and if so, how do they compare to traditional methods?
I'm aiming to enhance my video collection without compromising on the original quality.
Thanks for your help!

NickShaw is a spammer!!! And concretesrichmondva is a spam site!!!! Send your spam to concretesopinthechat2@gmail.com

[rgr]

If you have a well-denoised video, preferably saved as FFV1, you can try Topaz Video AI. It should be able to handle something like that. But I don't know if it's worth it.

[Selur]

Are there specific settings or filters that you find effective in preserving detail while minimizing noise?

Denoising always requires adjustments per source (sometimes even scene), since denoisers just can guess what noise and what details are. (not even thinking about 'wanted/intentional noise' that should be kept)

I'm aiming to enhance my video collection without compromising on the original quality.

Problem is that no program knows what you would count as 'original quality' compared to 'noise' or 'artifacts'.
Machine learning based resizers:
a. usually basically resize and sharpen (contrast enhance) and then run into the problem that they might enhance noise and lessen the 'original quality'
or
b. additionally, try to denoise and guess what could be noise, thus changing the content potentially a lot and removing parts of the image you would say are part of the 'original quality'.

What sort of content are we speaking of?
What have you tried? Why are you not satisfied with the outcomes?

If your source has specific characteristics and you can share a clip folks might be able to suggest or showcase what some of the tools might output.
But in the end you need to decide what is good and what not.

Cu Selur

Ps.: For clean sources NNEDI3 resizing + CAS is often enough 'enhancement', no fancy machine learning needed.

[pandy]

https://upscale.wiki/wiki/Main_Page
https://openmodeldb.info/

[Darryl In Canada]

For 1080p content from a decent camera, I think that Topaz Video AI does a nice job. I wouldn't use it for VHS-type sources. But anything from MiniDV upwards tends to be nice. You can test it out on the free watermark version. Lots of settings to adjust as needed. Topaz has a very active forum where users share settings and best practices.

The main alternative is to use the free tools either as scripts or more conviniently as housed in Hybrid. Hybrid has the tools to match TVAI, but the interface of TVAI makes it more intuitive to fiddle with the settings.