You need to make the pixels more dense before upscaling with wonder 2.
Yesterday i did enlarge a Phase One IQ3 150 image back to its 15k size after downscaling it to 4.5k, via wonder 2, it did look better than the original image afterwards, but not in every place.
My sweet spot is 3.1x, you downscale your image with Bilinear scaling (by 3.1x) then scale it up with wonder 2.
Thanks, here I increased pixel density within GP from 300 to 600ppi, is that what you meant? I don’t see a downsizing option.
No, you did get me wrong.
Load your image into Photoshop (or other editing software).
Downscale your image by 3.1x via Bilinear scaling.
Load the small image into Gigapixel, scale it up by 3.1x. (you will expierience the scaling bug here, your image will be cropped).
Or if your image did become too small for good upscaling after downscaling, use a different down scale like 1.59x and then scale up by 3.1x.
With my Eos R3 images, i go from 6000px down to 3766px then up to 11674px with a very good result.
Something like this… I can’t see doing this math each time, PITA 
. Unless there’s some quick way to input the resize, please provide a screenshot.
of?
Yes, thats fine, seems like the image does need a little bit of denoising before downscaling.
How do you determine the downscaling figures for each particular image? Or do you have an action that does it for you?
If the reduction amount if simple to input into the PS Image Size dialog, please advise. I routinely downsize by inches/resolution, but never percentage.
Perhaps if you are working only with the EOS images, they will all get the same treatment. I have tons of various size film scans and different resolution digital images spanning decades…
At the moment the targetdownsize for my cameras is 3766px.
At 3766, my cameras have the pixel size of a medium format camera with approx. 100 megapixels, which was my test size and proved to be very good.
It would be possible to write an action for this and apply it to a batch using automation.
I’m just not sure whether Gigapixel can be integrated into automation, but once you’ve reduced all your images to a uniform size using Photoshop or Camera Raw—a size that you’ll have to determine yourself as the sweet spot—then you’ll have come a lot further.
You can simply load images without layers into Camera Raw and reduce the size of all images at once. Of course, you should choose a file format that does not lose any details.
Thanks again for the tips and for doing the initial research!
Here is a picture made with the R5.
It may be that reducing the size destroys details, which we don’t want to do. We just want to bring the details closer together so that the Wonder 2 model can do its job.
For this camera, I will likely need to determine a resolution between my sweet spot of 3766px and 6995px.
The original resolution of the R5 is 8192px.
This is all very good information! But I am struggling with the practicality of doing this for every image you might edit…
Didn’t one or more of the Topaz apps have an option for downscaling before upscaling? I am not seeing this any more. And wouldn’t a workflow such as you describe be easier to use if it was baked in?
Recovery 2 has that.
It would be cool if they incorporated that into Wonder 2, but with the option to select downscaling so that it reduces the size with the same color depth and file format (not that it works with compression internally) and perhaps includes other methods besides bilinear for reducing image size.
Not necessarily, images can be quickly reduced in size in bulk.
I used a target resolution here and not a factor, maybe that was a mistake.
I use 3.1x (sweet spot so far), which was not the case with the 6995px image, where I typed in 11648, which alone can change the result.
I did send you an Message (DM).
In that ‘swimsuit girl’ example I posted there’s a visible pattern of what’s possibly destroying the details in Recover v3 (it’s some sort of mesh pattern), as the source image is lowres.
But you have to admit, his question is really silly. The answer is so obvious that I figured he was either doing it on purpose, or he wasn’t thinking straight, or he just lacked logic. Because it’s so logical: “the more files you add, the bigger the software gets.”
It was a question that shouldn’t even have been asked. Hence my very direct answer.
For me, it’s the kind of question that’s annoying because of the overly obvious answer.
This can happen when it’s overprocessed. Maybe that image would benefit from another model. Our models are not universal, but I can say confidently that we received a lot of good results from Recover 3
Or maybe the English language is not his native tongue and he was struggling to get his thoughts across better.
Regards 
Crop bug was fixed.
Now that did change everything.
My new upscale factor is 4x, 3.1x looks bad (artifical) now.
Thanks for the update!
Hum, So, there are translators powerful enough to do the job. English isn’t my first language either, so I use a translator.
