[biz-engineer]

In a search to improve perceived quality in large prints, I've tried various combinations of up-sampling and sharpening, including Gigapixels AI.

Gigapixels AI was overdoing it, the introduction of fake textures wasn't what I was looking for. At the same time, multi-pass sharpening and masking with conventional tools was time consuming.
Finally, I tried Topaz Sharpen AI as the final stage for the sharpening before printing, still using conventional tools for up-resing to 300ppi before going through Sharpen AI stage. Then I decided to buy Topaz Sharpen AI license, because I found it well justified, I can share the reason here below:

1) Sharpen AI is able to automatically distinguish between areas in focus and out-of-focus areas (using image feature recognition from its database, I suppose), so that it can even reduce noise in out of focus areas and improve visual acuity in focus area at the same time.
2) I've found that it can correct some type of lens uneven sharpness in corners (it correct D-FA 28-105 corners at 28mm), which allows to push image enlargement further.
3) Sharpen AI is slow to process, but the processing doesn't take effort of me, efforts are from the computer and processing can be batched/back-grounded.
4) Relatively inexpensive (especially if using a coupon-code, sometimes available via PF).

One of the by-product of using Sharpen AI was that I discovered that a lot of my images have some kind of blur:
- blur from shutter speed/camera handheld, or mirror/shutter vibration
- often blur due to not being perfectly focused

Which led me to the obvious conclusion (please don't laugh...), that exposure technique must be perfected for doing big prints:
- Tripod
- MLU / Electronic shutter
- Manual focus with LV magnification + DoF simulation

All this is not always possible/convenient.

And lens corner sharpness is the limitation, not sensor resolution, but can be somehow recovered with Topaz sharpness AI depending on the lens used.

---

Last edited by biz-engineer; 02-27-2021 at 01:02 AM.

[Paul the Sunman]

If you're using the 28-105 on the K1 at shutter speed between about 1/100 and 1/200, you might be seeing shutter shock. I don't see it with heavier lenses.

[Rondec]

If you're using the 28-105 on the K1 at shutter speed between about 1/100 and 1/200, you might be seeing shutter shock. I don't see it with heavier lenses.

I think he mentions using a tripod and the electronic shutter whenever possible. That typically should get rid of shutter shock (I never see it on my pixel shift images which use the electronic shutter).

[biz-engineer]

If you're using the 28-105 on the K1 at shutter speed between about 1/100 and 1/200, you might be seeing shutter shock. I don't see it with heavier lenses.

My images taken with the 28-105 are actually better than the ones from the 24-70 . My discovery was that Topaz Sharpen AI is able to correct the optically fuzzy corners of the 28-105 (at 28mm) almost entirely, and is much less able to correct soft edges/corners of the 24-70. I guess the "Stabilize" algorithm is able to detect gradients in an image, it interprets optical fuzziness in corner as camera shake and apply some local distortion to the image to correct it. I ran a pixel shift (dfa28-105@28mm) image through Sharpen AI, and it pretty much corrected the corners so it looks like the image was taken with a nearly perfect lens. But on another image that contained actual graphical gradient (wanted), Sharpen AI "stabilize" mode modified the image, so I chose not to use that "stabilize" mode but use the "sharpen" mode instead.

---------- Post added 27-02-21 at 13:44 ----------

I think he mentions using a tripod and the electronic shutter whenever possible. That typically should get rid of shutter shock (I never see it on my pixel shift images which use the electronic shutter).

I think I probably didn't express my thought properly in my post. What I'd say is that Sharpen AI made me realize how much better some images would have been if I had taken more care for the exposure (tripod, ES etc). Sharpening can correct / improve somehow, but the result is not guaranteed, it can also add unwanted artifacts or noise. Next time I'd be more careful with the exposures, which will allow me to sharpen less. It's just than when printing big, every detail is magnified, more care should be taken for the exposure.

---

Last edited by biz-engineer; 02-27-2021 at 05:44 AM.

[texandrews]

I had similar problems with GP AI, but managed to resolve them pretty well. Absolutely don't use the auto function at this time, and also be judicious about the size in crease---the 6x claims need to be taken with some salt, like maybe a 25kg bag. I found that it was really excellent at more modest amounts.

That said, I'm getting more interested in Sharpen AI and the denoise program. Now I see that they also have GP AI for video, which might have some use for me.

[biz-engineer]

I found that it was really excellent at more modest amounts.

Modest amounts of up-sizing is the reason I went for Sharpen AI instead of Gigapixel AI.

That said, I'm getting more interested in Sharpen AI and the denoise program.

IMO, all these tools are based on the same core image processing algorithms and AI database. Each tool is assembled into a software product with emphasis on noise reduction, sharpening, up-sampling etc.
The big value for me, for this software suite is that I can improve old photographs taken at a time when digital cameras weren't as good as now.

[texandrews]

IMO, all these tools are based on the same core image processing algorithms and AI database. Each tool is assembled into a software product with emphasis on noise reduction, sharpening, up-sampling etc.
The big value for me, for this software suite is that I can improve old photographs taken at a time when digital cameras weren't as good as now.

I think that's a perceptive take on it---I'll wager you are correct.

The value for me is different:

1. The ability to get the biggest possible images for work purposes. That's why I bought GP AI in the first place, for museum display support. With the 645Z and GP AI I was able to get outstanding prints over 48 inches on the long side, effectively 1:1 such that the documents placed on top of the print matched in size and quality the representations of them in the print. These were displays of Duchamp's White and Green Boxes' contents, which had to be rotated out individually due to light sensitivity. I'm sure there will be more uses once we're out of covid for various graphics purposes.
2. The ability to get my current photographic artwork enlarged w/o loss of quality (when they remain as pure photographs). I don't need these images to be larger than what GP AI can give me, as they are ganged up in multiple images and there's a limit to what I can practically handle in a framed work that has multiple prints in it (think diptychs and triptychs).

My older work is for the most part better as much more modestly sized images. GP AI will help a bit there, too, but is less necessary.

[pschlute]

still using conventional tools for up-resing to 300ppi before going through Sharpen AI stage

What camera are you using and what is your intended print size ?

Sharpen AI is able to automatically distinguish between areas in focus and out-of-focus areas (using image feature recognition from its database, I suppose), so that it can even reduce noise in out of focus areas and improve visual acuity in focus area at the same time

I do this myself in Photoshop with selective sharpening/NR.

that exposure technique must be perfected for doing big prints

Absolutely agree.

[DeepThrob]

big prints are time consuming

[biz-engineer]

What camera are you using and what is your intended print size ?

Print size, 24x36" (60x90 or 60x80cm). I gave up printing 30x40" from K1 files, at least not images taken with a zoom lens, I think such camera isn't meant to print that large.
But, it depends. Subject + blurred background can print large from full frame K1, crisp center can be printed down to 150ppi without being soft.
Let say, I had bad experiences printing wide-angle (all in focus scenes) from apsc files (DA15ltd), usually pretty bad at edges & corners, very visible.
IMO, printing at 300ppi native produces good quality prints reliably. From 200ppi and below, lens characteristics start to become visible.
I've checked out Yarrow's (David Yarrow) gallery enlargements from his D810 and D850, but he's got the trick: he uses the very best primes lenses from Nikon, 35mm f1.4 -> 105 f1.4 and fill the frame with subject, as close as possible, so he's able to print 60" large from full frame. Something to learn here.

---------- Post added 28-02-21 at 08:06 ----------

big prints are time consuming

:-) You are right. Also much more rewarding than 27" computer display.

[Viking42]

1) Sharpen AI is able to automatically distinguish between areas in focus and out-of-focus areas (using image feature recognition from its database, I suppose), so that it can even reduce noise in out of focus areas and improve visual acuity in focus area at the same time.

So this is an interesting comment. Can one assume then, that masking out-of-focus areas to selectively sharpen only in-focus parts is not necessary using this software? It seems from what you're saying is that it achieves the same goal without the extra work.

I'm using a K5ii at present, and this program is on my radar to buy. Sounds promising.

[biz-engineer]

So this is an interesting comment. Can one assume then, that masking out-of-focus areas to selectively sharpen only in-focus parts is not necessary using this software?

It's possible, I've done it, but time consuming, if I also consider other steps like merging exposure bracketed images or even merging bracketing panoramas... in the end, more processing step is more time spent in PP. I prefer to have a two step process: adjust images for color/light and automatic sharpening as the last step before printing.

[Viking42]

It's possible, I've done it, but time consuming, if I also consider other steps like merging exposure bracketed images or even merging bracketing panoramas... in the end, more processing step is more time spent in PP. I prefer to have a two step process: adjust images for color/light and automatic sharpening as the last step before printing.

Sorry, I think my question was too vaguely worded. My bad. What I meant to ask is whether Topaz is so good at recognizing out-of-focus areas that you don't have to selectively mask and sharpen the in-focus bits any more? The latter is what I have been doing for certain images, and, as you say, it's time consuming fiddly work. It would be great if Topaz did that for me, and sharpened only in-focus stuff without masking. Just wondering what your experience has been with that? From your description it sounds like it does it rather well.

[biz-engineer]

Topaz is so good at recognizing out-of-focus areas that you don't have to selectively mask and sharpen the in-focus bits any more?

There is a fully automatic mode, and a mask mode such that user can draw a mask to include / exclude areas for sharpening. The default mode does a good job, but it using mask can be beneficial for some images e.g image with mixed cloudy sky capture with long exposure (blurred clouds), the software doesn't quite know how to handle the blurred clouds because there is some details and some areas smoothed depending on cloud motion; in that case it's better to draw a mask to not apply AI sharpen to the sky. In photos with clean subject separation (subject in sharp focus, and blurred background e.g flowers close-up, bird photos with long lens etc..), the default Sharpness AI mode (auto) works very very well.

[Viking42]

OK -- got it! That's really good to know. Thanks for the reply - very helpful. Good to have insight like yours into these kinds of products.

Best,
Svend