Ai Upscaler/Enlarger (GPU and CPU based)

[Katsumi]

Enlarge images (like 2 times) is with XnView Classic/MP blurred and in most cases (for me) not usable. Its unfortunately true for mostly all graphics tools like GIMP. XnView is a very great tool and have nice converter options, however shouldn't it also be able to enlarge images in a modern way?

There exists tools since many years that do this work excellent. The most avatars in communities have cartoon-look, so here open source ai upscaler specialized for cartoons as example: https://github.com/nagadomi/waifu2x or https://github.com/bloc97/Anime4K
Unfortunately this sources are poorly maintained or stalled.

However I understand when this is not what XnView want to offer in this app. I just wanted to throw it in. There is a open source windows app thats have it https://github.com/AaronFeng753/Waifu2x-Extension-GUI

[helmut]

Welcome to the XnView forum and thank you for your suggestion, Katsumi. Including a smart upscaler/enlarger into XnView MP would be indeed a cool thing. Somehow I have in mind that there is a plug-in available for XnView Classic which does smart enlarging but I guess I'm wrong.

Enlarge images (like 2 times) is with XnView Classic/MP blurred and in most cases (for me) not usable. Its unfortunately true for mostly all graphics tools like GIMP. XnView is a very great tool and have nice converter options, however shouldn't it also be able to enlarge images in a modern way?

You may try out enlarging using "Resampling: Nearest Neighbour". XnView will just multiply pixel and image is enlarged without blurring. But: Your image will look pixelated because existing image information is just copied/multiplied and no new image information is created/calculated using a smart algorithm. So I guess that's not really what you want but I wanted to mention this possibility.

Related topics with good information and links on this matter:
- Image enlarger/Magnification filter
- resizing algorithms

[Katsumi]

Currently I using https://github.com/nihui/waifu2x-ncnn-vulkan that can used only with CPU or GPU. Its have converter models for photo and cartoons. Its slow, but very good results. No AI training or external connection needed.