[JohnMc]

... but my reading of his original post is that it's things like the false textures that Gigapixel can produce that he's concerned about.

Is it really "false" or just bad or lesser re-interpretation(s) of? It's possible to work with multiple outputs from GP and blend as layers to try and dampen the undesirable effects variable guesses as to what the whole cloth should look like if only it were bigger.

[Dartmoor Dave]

Is it really "false" or just bad or lesser re-interpretation(s) of? It's possible to work with multiple outputs from GP and blend as layers to try and dampen the undesirable effects variable guesses as to what the whole cloth should look like if only it were bigger.

That's an excellent point. For myself, I use multiple layers exactly as you suggest. Usually Lanczos for the bits of the photo where that works and Gigapixel where it works. Reality is that which existed independently of what we thought about it when we took the photo. A truthful photo is one that comes as close as we can get to the thing that was real.

[JohnMc]

... as close as we can get to the thing that was real.

Maybe I'll go back to one of my Stereo Realists then. ;

[Dartmoor Dave]

Reading that "nature of reality" stuff again after eating dinner, it occurs to me that I might need to adjust my meds.

Edit:

Although. . . I actually do think there was a stereo (solid) reality there when I took the photo. I don't know what that reality really was, but there are various ways that I can justify saying that my photo of it has got a certain amount of truthiness to what it might have been.

Yep, definitely need to tweak the meds.

---

Last edited by Dartmoor Dave; 06-15-2022 at 12:37 PM.

[JohnMc]

It used to be fun when I got kids from criminology over to do their gen ed elective with me and I'd introduce them to the works of Les Krims, Duane Michaels and of course Ernest Bellocq. I was always surprised at how many would be offended and want to drop. They'd usually come over in in groups of 3 or 4 and 1 or 2 would complain and leave. Yes, it was partly shock, but always intended to further discussion or open eyes and mind to those other realities we have. It was amazing that they would be offended by such work, granted not everyone's cup of tea, yet wanted to be criminologist.

[Dartmoor Dave]

It used to be fun when I got kids from criminology over to do their gen ed elective with me and I'd introduce them to the works of Les Krims, Duane Michaels and of course Ernest Bellocq. I was always surprised at how many would be offended and want to drop. They'd usually come over in in groups of 3 or 4 and 1 or 2 would complain and leave. Yes, it was partly shock, but always intended to further discussion or open eyes and mind to those other realities we have. It was amazing that they would be offended by such work, granted not everyone's cup of tea, yet wanted to be criminologist.

I'd never heard of any of them (except perhaps I've seen some Bellocq somewhere along the line). But I promise I won't drop the course, and thanks to Google Images I now think that Les Krims might soon be one of my all-time favourite photographers.

Don't want to be a criminologist though. Never could stand the sight of blood.

(Speaking of criminology, sorry about the thread hijack Mike.)

[GUB]

Don't you just love the movies where a satellite image is "enhanced" and then "enhanced" again onwards until they can see what brand cigarette the baddy is using!
Sorry but new tech still hasn't debunked the good old computing law:
Garbage in - garbage out.

[JohnMc]

... might soon be one of my all-time favourite photographers.


(Speaking of criminology, sorry about the thread hijack Mike.)

He will give you a different appreciation of chicken soup. Ditto to Mike, who didn't need GP to become BigMack. At least I hope not ;

[dms]

The comments (posts) seem to center on two areas: 1. software is useful but problematic as it/they are altering the images and 2. like anything you need to take care and check it.
These are actually very divergent as there is a difference between (for example) sharpening too much and getting halos and sharpening and getting a sixth toe.

One can know how usual sharpening works and thus what to look for, but one does not know what the software in question does. That is a big difference. For all I know it will replace eyelashes with swastikas, but if I don't check everywhere at the pixel level I cannot be sure!

[JohnMc]

...difference between (for example) sharpening too much and getting halos and sharpening and getting a sixth toe.

For all I know it will replace eyelashes with swastikas, but if I don't check everywhere at the pixel level I cannot be sure!

Angels or monsters. It's akin to hitting the road with Quinn Quicksilver, Gustave Dore, or Howard Finster. Love it. Your later point is of course true with any bit of code, be it in MS Word or the link in the text message you just clicked.

[BigMackCam]

Thanks for the interesting responses, all I fully expected there'd be a mix of support, push-back, hints at the importance of user (i.e. "my") abilities, and just general good discussion, as is usually the case

In another thread from today, fellow forum member @biz-engineer suggested that the views I'd expressed here left an "after-taste" of being "unsatisfied, frustrated, annoyed" with the Topaz AI tools. I want to clear that up, because if that's the general impression I've given, I've failed in the way I presented my views. For the record, I think these tools are very clever, can produce excellent results (depending on image content and quality), I find them useful, and I'm glad I purchased them - all three of them, actually, as I've been able to rescue and/or improve upon numerous photos with each one in ways that the others couldn't, depending on the primary issues - whether that's overal softness, noise, pixel dimensions and/or JPEG compression artefacts. This thread was never intended to be a rant for me to vent my frustration and annoyance with these tools, because (a) I've never felt frustrated or annoyed by them, and (b) it's not typically in my nature to rant Actually, I've found it very interesting to learn how to use them, and - I repeat - I've improved a number of my photos with them; so I'm very glad I bought them, especially at the heavily discounted price I paid. Worth every penny, no question.

My trust issue is essentially to do with inconsistencies in content, textures and overall results from certain AI models (sometimes only at certain adjustment levels) that I wouldn't ever see with pixel-level raster processing. With my pre-AI workflow, the only reason I would need to carefully examine every area of an image at 1:1 reproduction in detail is for optional spot removal (litter, birds in the sky, sensor dust spots, that kind of thing). Otherwise, I can unfailingly trust my mostly global adjustments to tone, noise, sharpening and other aspects to work across the entire image quite effectively. With AI tools, I feel there are enough instances of unusual / unexpected results - "sixth toe" or "dodgy ear" as extreme examples, but also more subtle issues such as very different noise reduction textures in different parts of an image, or strange-looking text - that I have to check carefully, and accept that in some shots (not an insignificant number, based on my current level of experience) I may have to perform further raster editing - and that editing may not be entirely straightforward depending on the quality of result I'm looking for. Indeed, as one or two of you have pointed out, multiple processing cycles may be necessary to obtain the desired result.

I hope - with that explanation - there's no ambiguity remaining, and no room for interpretation beyond my original intentions. Again, the purpose of this thread wasn't to rag on AI tools, but to point out why I personally don't trust them yet. That's all, nothing more. I hoped that my original post and members' responses might be educational to others who may be considering these tools, and helpful in setting expectations - and I hope that's still the case.

Thanks again for the responses, everyone, and to @biz-engineer for his feedback in the other thread

[Michail_P]

Enormous distortions in judgement of imaging these days. People zoom in at whatever magnification is possible, 100%, 200%, 300% , completely losing sight of how the image would look like as a paper / canvas print at its final size. The wrong judgement: never sharp enough at 200% on electronic displays, and yet never make large prints (or never make prints at all..).

Time to plan and travel to new location, use the camera, get off the internet hook.

Couldn't agree more. Sadly many prefer peeping to admiring the artistic result or the actual captured subject / spectacle.

[Rondec]

I think the issue is whether computational software can replace a dedicated camera (not a smart phone) with a skilled photographer running it. The answer for many people is, "Absolutely." They understand the limitations that such gear has, but they are more than willing to deal with those for the convenience they get with a camera on their phone.

Adobe and Topaz's software do amazing things. They do not, of course, make a boring photo interesting. They do not improve compositional issues. While they may "create" detail to let an image be enlarged, that detail isn't necessarily accurate. Personally, I mainly use a 36 megapixel camera for the images I make and even cropping moderately, that is plenty for the size images I print. Considering that few people print at all these days, I am not totally sure how much call there is for software that turns your 36 megapixel image into a 100 megapixel image. Probably more than I realize, but the sharpening and noise reduction packages would be far more useful to me than the Gigapixel AI.

Of course, I'm not everyone and so my needs may not mirror those of the wider community.

[biz-engineer]

Probably more than I realize, but the sharpening and noise reduction packages would be far more useful to me than the Gigapixel AI.

The first thing that shows in photographs is lens aberration, the second thing is aliasing/false colors, applying G.P. A.I on top of an image having non-perfect edge is not a good idea. CA correction on RAW is not perfect, DXO PL does a better job than DCU or Silkypix, depending on the lens.

Then come the pixel PPI thingy. Printing at 24x36" (small poster) requires 300 PPI files ready to send to their printers, that's >70Mpixels, up-sampling soften the image, plus the print process also soften the image output (depending on paper and spread of ink dots or RGB laser spot glow on silver paper), a good up-sampling & sharpening tool makes a visible difference on print output. Now, who print at 24x36"? I'd say, the ones who used to shoot medium format film are good candidates, and the ones who used to shot 135 film rolls don't need it. For me Topaz sharpening is a game changer, as I'd probably be shooting Fuji GF already if I didn't have Topaz Sharpen AI and K1 pixel shift (pixel shift is actually very good when up-sampled).

[bdery]

they give you is content replacement rather than enhancement

That's a good way to put it.

they're useful to have in my post-processing toolbox. With careful use - and, often, some manual editing after the fact - they can work wonders... but sometimes even the most careful use can result in unwanted artefacts or unusual results.

Agreed again.

I think part of the problem is a dialogue of deaf between people who think photography should represent , document the world, and those who feel that photography is art and everything is fair game.

I see dangers in blindly following the second, but I see unnecessary limitations in doing only the first.

I tend to dislike anything close to a deepfake, and in this POV I don't really like replacement.

I am not pro or con Tony. He is honest yet often not informed enough.

Not being informed enough accounts to being wrong.

Having an agenda brings inaccurate and misleading opinions and "facts".