Photos, made with AI

[+PISA-caching]

Lately I started experimenting with the new Beta version of Adobe Photoshop. With it I was able to expand a photo of a fountain and let Photoshop fill the empty space around it. The result is stunning (if you ask me). Judge for yourself:
 

 

Left is of course my original photo. I created a WM for that fountain and of course I used only the original photo. But what do you think? Would it be ok (in your opinion), to cut out - let's say - some dirt or other distracting things and let Photoshop fill the hole? And where's the limit then? One limit I could think of is: The motif of the photo has to stay untouched, meaning: To expand or modify the motif of the photo (like I did above) is not ok, pimping the surroundings is ok, as long as it doesn't completely change the look. I don't plan to use AI to modify my photos, but I'm eager to hear what you think about it.

Edited July 28, 2023 by PISA-caching

[Country_Wife]

Well… I don’t have Photoshop, but on some recent photos there were distracting patches of bright colour. I used a very crude photo editor to dull down the colour. I didn’t add or subtract anything to the main subject, just reduced the distracting elements. All of photography, going back to the darkroom days, has been about enhancing the original image in various ways: push developing of film, burning and dodging different elements of the print… As long as the alterations are artistic, but the photo is still true to the main subject, I don’t have a problem with photo editing on waymarks.

Edited July 29, 2023 by Country_Wife

[+ScroogieII]

On 7/28/2023 at 9:53 AM, PISA-caching said:

Lately I started experimenting with the new Beta version of Adobe Photoshop. With it I was able to expand a photo of a fountain and let Photoshop fill the empty space around it. The result is stunning (if you ask me). Judge for yourself:
 

 

Left is of course my original photo. I created a WM for that fountain and of course I used only the original photo. But what do you think? Would it be ok (in your opinion), to cut out - let's say - some dirt or other distracting things and let Photoshop fill the hole? And where's the limit then? One limit I could think of is: The motif of the photo has to stay untouched, meaning: To expand or modify the motif of the photo (like I did above) is not ok, pimping the surroundings is ok, as long as it doesn't completely change the look. I don't plan to use AI to modify my photos, but I'm eager to hear what you think about it.

 

Interesting that you should address this topic, Andreas.

Not having access to more advanced Image Editing software (IE, as opposed to AI ), I have for years been, whenever it appeared necessary in the interest of improving Waymark pix, doing essentially what you presented here. Removing a passing pedestrian (or motorist), extinguishing my shadow, extending roadways and sidewalks into trashy, weedy land, removing the weeds and grass from sidewalks, plazas and roadways, extending background into the sky, removing sky to provide more background, eliminating or mitigating glare and flares, relettering building signage from old pix to represent their current status, layering "then" & "now" pix to create "then-to-now" and "now-to-then" transitions for the eponymous categories - I've done them all, and more. I've removed hundreds & hundreds of power and communications lines from pix. All that I've done mostly manually.

 

And, finally, I have always also enhanced my pix, improving lighting, contrast and/or saturation whenever I deemed it necessary. I review a LOT of waymarks lately and, it seems that cameras haven't improved much in the past decade or so with regard to their light capturing ability, or better stated, their ability to reproduce the light capturing ability of  the human eye. Essentially ALL pix I review would be improved, sometimes markedly, with just a touch of enhancement. 

 

So, my take on this would be that anything which improves, or ameliorates, the aesthetics of a pic or removes, or mitigates, distractions from the subject is to be encouraged.

 

But now, as usual, Andreas, you've given me something else to ponder: Just when does clever coding evolve into AI? In the case of Photoshop I think not, as AI involves reasoning, self learning and self teaching, which I doubt takes place within Photoshop. I imagine that it entails no more than a bit of "clever coding". Not being well up to speed on AI, that's about all I can say on the matter. But thanks for tickling a few brain cells!

 

Just a second before we go!

I again looked at your and Photoshop's side-by-side pix, this time more closely. I see that Photoshop was forced to make a couple assumptions in its expansion of your pic. In the foreground it added a few small clumps of flora in the cobblestone, which would have required an assumption on its part that said flora would not be out of place, given that flora was seen in your pic. However, it made a rather poor assumption with regard to the pattern of the cobblestone on the far right background, laying it in essentially straight lines, which appear nowhere else. This bespeaks clever coding, certainly not AI. Moreover, the boards on the near, long side of the fountain run full length in the original, whereas the boards on the left end of the Photoshoped version are continued to the end, making the boards on the far, long side of the fountain NOT full length. Again, an incorrect assumption, unless carpentry there employs an unusual aesthetic.

Keith

Edited July 29, 2023 by ScroogieII

[+PISA-caching]

12 hours ago, ScroogieII said:

In the case of Photoshop I think not, as AI involves reasoning, self learning and self teaching, which I doubt takes place within Photoshop.

 

Photoshop generates content for the empty pixels that I added to the left, bottom and right of my original photo. It not just generates one version, but three and I can judge every version with thumbs up or down. That's the learning process for Photoshop. Thousands of Photoshop users do that all the time and Photoshop is constantly learning, what is good or bad. When I made the photo above, one version contained a big terracotta vase on the right side :-), which of course received a thumbs down. And it took about 20 versions before one was generated that received a thumbs up by me.

 

I agree that the result isn't perfect, but there are two main reasons for that: 1. This is a Beta version, that didn't have much time to learn so far. 2. I'm not at all a specialist in Photoshop and with more Know-How and time I'm quite confident that I would have been able to remove/avoid the errors you mentioned. In fact, I was not really satisfied with the front corner of the stone base of the fountain and removed a small part and let Photoshop again fill it, until I was satisfied. And Photoshop added some kind of table or something to the top right corner, that I would have removed if I had more time.

 

Apart from all that, I agree with both of you, that improving the lightning, removing trash or passengers etc. is ok. At least I could have removed the trash myself before taking the photo or wait until the passenger is gone. I did all that myself and I used a Canon software to combine several of my photos to a panorama photo. It's just that I think that adding completely new content to a photo is questionable, when it changes the main motif of the photo. Here's another photo I increased to the left and bottom:

 

One of the generated versions added a small tower on the back side of the church. I think we all agree, that this is not ok. Also, I'm pretty sure that the indicated columns on the left side of the building reach from the roof to the floor in reality, which they don't on the Photoshop image.

 

I'm just curious, where all this will lead. Will there be times, where I can download a photo from the internet and tell Photoshop "change the point of view 20° to the west and make the shadows look as if the photo was taken on the afternoon" etc. This would make it almost impossible for officers to tell, if the photo was stolen from the internet or not. 

 

 

[+Bear and Ragged]

5 hours ago, PISA-caching said:

I'm just curious, where all this will lead. Will there be times, where I can download a photo from the internet and tell Photoshop "change the point of view 20° to the west and make the shadows look as if the photo was taken on the afternoon" etc. This would make it almost impossible for officers to tell, if the photo was stolen from the internet or not. 

 

Given Google Street View (and others) and the availability on-line of images of the more popular tourist haunts, I feel this is more 'When' rather than 'If' it will happen.

[+ScroogieII]

7 hours ago, PISA-caching said:

I'm just curious, where all this will lead. Will there be times, where I can download a photo from the internet and tell Photoshop "change the point of view 20° to the west and make the shadows look as if the photo was taken on the afternoon" etc. This would make it almost impossible for officers to tell, if the photo was stolen from the internet or not

 

As Bear and Ragged noted, the "if" is quite likely in our rearview mirror. I couldn't be certain about it, but I suspect that the version of Photoshop you're playing with right now is capable of changing the point of view of an image, especially that of a spatial, three dimensional object such as a building. AAMOF, I have been able to simulate minor changes in point of view with GIMP, manually, of course. If you review either of the "Then & Now" categories, you may have encountered an example and not noticed my handiwork. 

 

As for other capabilities of your version of Photoshop, I'd say it's pretty impressive, but still has a bit of improvement in store.

 

Your description of Photoshop "learning" capabilities, seems just a bit "off" to me. Photoshop itself doesn't seem to me to be learning anything, just offering a random selection of renderings from which you choose the best, at which time it generates yet another random set of renderings. It's unclear to me whether each generation of renderings always becomes more acceptable than the previous generation. Does it?

Keith

[+PISA-caching]

3 hours ago, ScroogieII said:

Your description of Photoshop "learning" capabilities, seems just a bit "off" to me. Photoshop itself doesn't seem to me to be learning anything, just offering a random selection of renderings from which you choose the best, at which time it generates yet another random set of renderings. It's unclear to me whether each generation of renderings always becomes more acceptable than the previous generation. Does it?

 

Well, I assume (but haven't found any prove of my theory yet), that my decisions (thumbs up or down) are reported to Adobe and that that will help Photoshop to give better results in the future. Whenever I click on thumbs up or down, a small overlay appears saying "Tell us more" (in German as I use a German Photoshop) and when I click on that I can give further information. I can rate 4 statements:

 

1. I would use that result for my project

2. The result fits to my text input.

3. The result looks credible.

4. The result fits well with the rest of my picture.

 

All these statements I can rate with "Disagree completely", "Don't agree", "Neutral", "Agree" or "Agree absolutely". And I can add a text note, which someone (or the AI?) will read. So, I think that this will not just help MY copy of Photoshop, but all the Photoshop user out there. I learned that if I use "Generative Fill", the original photo is sent to Adobe, processed there and sent back to me. All that makes me believe that this is (at least in a way) a learning software. 

[Country_Wife]

Sounds to me like you’re training the AI! Also, Adobe says this feature is “powered by Adobe Firefly generative AI”. 

 

https://www.adobe.com/ca/products/photoshop/generative-fill.html

[+ScroogieII]

23 hours ago, PISA-caching said:

Well, I assume (but haven't found any prove of my theory yet), that my decisions (thumbs up or down) are reported to Adobe and that that will help Photoshop to give better results in the future. Whenever I click on thumbs up or down, a small overlay appears saying "Tell us more" (in German as I use a German Photoshop) and when I click on that I can give further information. I can rate 4 statements:

 

23 hours ago, PISA-caching said:

I learned that if I use "Generative Fill", the original photo is sent to Adobe, processed there and sent back to me. All that makes me believe that this is (at least in a way) a learning software. 

 

Been thinking about this for a day now.

OK, now that I'm a bit more up to speed with the flashy new incarnation of Photoshop, and assuming you are correct in your assumptions and beliefs, that does give rise to yet another question:

In what way would Photoshop's learning to correctly render cobblestones around a fountain (which it appears not to have learned) help to improve its ability to correctly render pilasters on a Byzantine church (an example of which the one above is not, BTW )?

Keith

[+PISA-caching]

8 hours ago, ScroogieII said:

In what way would Photoshop's learning to correctly render cobblestones around a fountain (which it appears not to have learned) help to improve its ability to correctly render pilasters on a Byzantine church (an example of which the one above is not, BTW )?

 

 

I don't think that the learning process is that fast, that one thumbs up or down with change anything, but other than that: If I were able to answer that question thoroughly, I would code the next version of Photoshop. 

[+ScroogieII]

Well, speed isn't really what I was referring to. It just seems to me that things learned from a successful generative-fill on one object or subject wouldn't, in most cases, be useful or applicable to another, quite different object. How could AI infer any useful relationship between cobblestones and pilasters, for example?

Keith

[+PISA-caching]

I quote myself:
 

On 8/1/2023 at 7:28 AM, PISA-caching said:

If I were able to answer that question thoroughly, I would code the next version of Photoshop. 

 

As a total idiot on AI matters, I would say that several (positive and negative) answers to cobblestone photos will (if at all) improve further cobblestone photos and nothing more.

[+ScroogieII]

Agreed.

[+FamilieFrohne]

On 8/3/2023 at 11:46 AM, PISA-caching said:

As a total idiot on AI matters, 

 

AI = artificial idiocy? (SCNR)

[+ScroogieII]

On 8/7/2023 at 11:29 AM, FamilieFrohne said:

AI = artificial idiocy? (SCNR)

 

Lovely, Erik!!! LOL

Keith

[+ScroogieII]

On 8/21/2023 at 12:23 PM, flucemma said:

the essence of a photograph is in its subject

 

Indeed! Though I may look into Mid Journey AI, simply out of curiosity, I, personally, am not likely to be putting it to use in future Waymarks, as I have, for the present at least, chosen to submit no further Waymarks. However, I shall continue to review submissions to the categories to which I have committed, possibly to eternity.

 

That said, I will again, as I have in many previous posts, encourage Waymarkers to make use of FREE image enhancement software to, at the very least, enhance photos provided to the extent that the operational aspects should become easily legible. That failing, clearly legible closeup photos of pertinent signage would, at the least, make the lives of reviewers easier.

 

A single commonality within this community is the submission of quite substandard photographs. While most are quite/somewhat legible (with regard to signage) or visible (with regard to buildings and other objects), a great many (in truth, almost all that come my way) completely fail to expose the true character of the subject. One or two distant or irrelevant photos of a sculpture, or a structure of a particular archaeological style, should, indeed, be unacceptable in such categories. Alas.

 

Improvement in the character and quality of a photo is extremely easily done - ofttimes the lighting levels, contrast, saturation, colour balance and other characteristics of a photo can be appreciably improved in less that 5 seconds. Believe it - I have been doing it for nearly 10 years. Face it, image enhancement software was developed purely because contemporary cameras remain incapable of capturing images nearly as accurately as can the human eye.

 

Beyond, that, given that we're (hopefully and trustingly) dealing with original photos taken by and submitted by Waymarkers, AI, in this particular context, is not likely to be our friend.

 

I am quite certain that, with the aid of (now) pedestrian software such as Faststone and Gimp, I could easily convince you that I had visited Machu Picchu this summer, which I haven't, though I certainly would like to.

 

So, NO AI isn't at all likely to be, nor become, the friend or ally of Waymarking, as its purview encompasses our dealing with REAL LIFE visits to REAL SITES, each extant in our REAL WORLD, each viewed and experienced in REAL TIME by REAL Waymarkers.

 

Keith

PS - That's certainly not to say that AI is valueless with regard to other aspects of our world, which it certainly is not. As just one example, I understand that AI is  currently being used to correctly present medical diagnoses to patients in areas where human surrogates (get the irony here? ) are unavailable.

Edited August 28, 2023 by ScroogieII