Though the picture that I used from long ago was exported from what were originally vector images.

Maybe a good test needs an image that is drawn chunky at a small size to start.

William

William

Anyone using an upscaler with AI ?

Well, as far as I know, I am not.

What I have done previously is that I wanted a hardcopy print of an image that I had produced about twenty years earlier and put on the web for something I was doing at the time. The picture is a combination of some Microsoft Clip Art and a few additions by me.

The final illustration in the following page.

http://www.users.globalnet.co.uk/~ngo/euto0008.htm

I enlarged the image, I think by three times in both x and y axes and placed it on a white background, got it printed as a custom greetings card, advertised as a photo card but as long as it is a jpg file it does not actually need to be a photograph.

https://www.papier.com/landscape-photo-313

I appreciate that an expert is art prints might well find it to be chunky, but it looks fine to me.

We can try to experiment.

I opine that the picture in this post would be good for testing enlargement.

https://punster.me/serif/viewtopic.php?pid=3876#p3876

This being because it has sufficient detail for such tests, yet is not detailed as much as, say, a picture of a group of many people.

If one clicks the picture then clicks the resulting picture then downloads, one can get a 1024 pixel by 1024 pixel image.

I am wondering if one makes a copy of that image and reduces it to a smaller size, maybe 256 pixels by 256 pixels, can one use some sytem, maybe AI based, to get back an image like the original?

As a more extreme test, what if a copy is reduced to, say, 128 pixels by 128 pixels, to what extent can a reasonable looking enlarged image be produced, sufficiently like the original picture?

What, one may ask and consider, do "reasonable" and "sufficiently" mean in this context.

William

Anyone using an upscaler with AI ?