Upscaling Question

[Jay123210599]

Which one has the better and higher quality, upscaling an image or upscaling a video?

If the former is true, is R-ESRGAN 4x+ the best upscaler for both normal and anime photos? What are some programs or websites that I can use it locally?

[poisondeathray]

Which one has the better and higher quality, upscaling an image or upscaling a video?

It depends on the upscaling method or algorithm used, and the source

IF you use single image neural net upscaler, there will be no difference, as only 1 frame is processed at a time, and the majority work in RGB anyways. Often a problem with single image upscalers when used on video , or an image sequence is flicker and temporal inconsistency

If you use video or temporal upscaler , they can make use of multiple frames and that can sometimes produce better results assuming you meant an image sequence, not a single image

If the former is true, is R-ESRGAN 4x+ the best upscaler for both normal and anime photos?

I would say probably not , but it depends heavily on the source and specific models used. It's one of those things you have to test and try. There are hundreds of models and upscaling architectures . Many of the open source ones use python

If your source is very close to the training set used to train the model, the results should be good.

But there are many different types of anime styles, many different types of "normal" photos. A single model for everything would look terrible. You need a specific model for the best results

The best upscales generally make use of pre and post processing appropriate for the source

What are some programs or websites that I can use it locally?

Chainner and cupscale can use many upscaling architectures, vapoursynth has access to many as well

https://github.com/chaiNNer-org/chaiNNer
https://github.com/n00mkrad/cupscale

[Jay123210599]

Which one has the better and higher quality, upscaling an image or upscaling a video?

It depends on the upscaling method or algorithm used, and the source

IF you use single image neural net upscaler, there will be no difference, as only 1 frame is processed at a time, and the majority work in RGB anyways. Often a problem with single image upscalers when used on video , or an image sequence is flicker and temporal inconsistency

If you use video or temporal upscaler , they can make use of multiple frames and that can sometimes produce better results assuming you meant an image sequence, not a single image

If the former is true, is R-ESRGAN 4x+ the best upscaler for both normal and anime photos?

I would say probably not , but it depends heavily on the source and specific models used. It's one of those things you have to test and try. There are hundreds of models and upscaling architectures . Many of the open source ones use python

If your source is very close to the training set used to train the model, the results should be good.

But there are many different types of anime styles, many different types of "normal" photos. A single model for everything would look terrible. You need a specific model for the best results

The best upscales generally make use of pre and post processing appropriate for the source

What are some programs or websites that I can use it locally?

Chainner and cupscale can use many upscaling architectures, vapoursynth has access to many as well

https://github.com/chaiNNer-org/chaiNNer
https://github.com/n00mkrad/cupscale

What are the best upscalers to use for videos and images?

[poisondeathray]

What are the best upscalers to use for videos and images?

For "neural net" or 'AI" category, usually the ones that were trained on sources that are most similar to your videos and images

You often have to use other filters in combination if you want best results . You almost never get the "best" results using 1 upscaling filter only

[Jay123210599]

What are the best upscalers to use for videos and images?

For "neural net" or 'AI" category, usually the ones that were trained on sources that are most similar to your videos and images

You often have to use other filters in combination if you want best results . You almost never get the "best" results using 1 upscaling filter only

How do you find the sources for blu-ray videos and movies, as well as video games (both original and remasters)?

[poisondeathray]

How do you find the sources for blu-ray videos and movies, as well as video games (both original and remasters)?

You buy the original movie and video game sources

If no free publicly available models exist in your category, then you can try to train custom ones for best results. Otherwise you have to use non specific models and conventional techniques and filters

There are various forums and discussion boards, discord for upscaling . Some have private models

Often you have to pre/post process sources (not all sources are pristine to begin with, not even BD)

[Jay123210599]

How do you find the sources for blu-ray videos and movies, as well as video games (both original and remasters)?

You buy the original movie and video game sources

If no free publicly available models exist in your category, then you can try to train custom ones for best results. Otherwise you have to use non specific models and conventional techniques and filters

There are various forums and discussion boards, discord for upscaling . Some have private models

Often you have to pre/post process sources (not all sources are pristine to begin with, not even BD)

Here is an upscaled Blu ray movie.
https://www.blu-ray.com/movies/Dragon-Ball-Super-Super-Hero-4K-Blu-ray/343894/

This is also a hd collection of remastered games.
https://slycooper.fandom.com/wiki/The_Sly_Collection

[poisondeathray]

Beware that sometimes a retail upscale isn't done very well . Often a studio will do a poor "rushed" job of an upscale . Sometimes you can find alternate methods and scripts with comparisons in various fan forums that sometimes do a better job

There are many game mods and upscaled textures for various games and platforms, just do search. There are forums dedicated to game upscaling. Many of the older upscales used ESRGAN derivatives, but there are newer engines now