| Moderated by: chrisbet, | Page:   ... 13 14 15 16 17 18 19 20 21 22 23 ...   |
|
|||||||
| Using AI in photography - Page 18 | |
| General discussion and graham's youtube film | Rating: 
|
| Author | Post |
|---|
| Posted: Tue Jan 16th, 2024 11:25 |
|
171st Post |
Eric
|
OK....looking again at this file info... Ive opened a new NEF in photoshop via camera raw and copied its file info metadata to the difference software. I then masked and applied an ai generated background then saved the file as a psd. Whist it was still on screen I copied its file info into the difference software. Ignoring a couple of lines that refer to the commencement of the activities which differ by 3mins ....the time to carry out the masking and the Ai activities, there are only 3 section that differ as follows.....  Click here to comment on this image.  Click here to comment on this image.  Click here to comment on this image. This seems to differ to the previous 'photometricinterpretation' reference example we pondered over. I confess to being baffled as I can see no reference here to 'Content Aware Credentials' on anything that tells me the right hand txt is from an ai altered image.  Unless Adobe haven't added it yet, I cannot see how an easily recognised reg flag can be thrown up for the average competition judge or even the average commercial artwork director who are unlikely to have programming know how to decipher the files metadata?
____________________ Eric |
||||||||
|
| |||||||||
| Posted: Tue Jan 16th, 2024 11:58 |
|
172nd Post |
Eric
|
Back to some AI enhanced images..... Click here to comment on this image.
____________________ Eric |
||||||||
|
| |||||||||
| Posted: Tue Jan 16th, 2024 13:07 |
|
173rd Post |
chrisbet
|
I reckon it going to become like trying to attribute paintings etc. - it is in the eye and opinion of the viewer to decide if an image has been subject to AI In the case of splendor in the grass, I would wonder how the two animals managed to get in there without trampling the grass?? There is an odd junction between the two animals. The DOF also seems a bit deep for a 200 at f2.8.
____________________ If it is broken it was probably me .... |
||||||||
|
| |||||||||
| Posted: Tue Jan 16th, 2024 14:31 |
|
174th Post |
Eric
|
chrisbet wrote:I reckon it going to become like trying to attribute paintings etc. - it is in the eye and opinion of the viewer to decide if an image has been subject to AI Well spotted, Chris! The junction between the cheetahs is an initial masking fault that I missed. Photoshop didn't see them as one joined object so I added them together and missed the fluffy bosom .....as did Photoshops object mask. Ive started overlaying the masked object over the AI layer as there are often weird artefacts at the edge of the mask (that's what you saw between them) but it merely overlayed the incomplete cutout.  Ive posted the corrected image, without embellishments, below. On the matter of 200mm @f2.8 dof .....you would probably be quite right if it were a full frame (ie the camera was nearer to the subjects) but it is a 50% crop, so you should expect to 'keep' the greater dof from the greater camera distance in the final cropped image.... that's my story anyway lol. The cheetahs came down the tree along a branch and dropped in. As my mother use to say...Ive "got a plaster for every sore"   Click here to comment on this image.
____________________ Eric |
||||||||
|
| |||||||||
| Posted: Tue Jan 16th, 2024 14:38 |
|
175th Post |
Eric
|
I hope the silent majority are benefitting from this warts and all exposure.
____________________ Eric |
||||||||
|
| |||||||||
| Posted: Tue Jan 16th, 2024 16:22 |
|
176th Post |
Graham Whistler
|
However, it was created it is still a good photo. I agree about the coding also most judges at any normal camera club meeting or RPS or BIPP event would not have a computer plus Adobe and original files as edited by the photographer. Even in a comp with projected PDIs? An art director in an advertising studio producing his work would be using the photographer's files to produce finished artwork etc. I think Adobe would need to add a more easy-to-find menu for policing such matters.
____________________ Graham Whistler |
||||||||
|
| |||||||||
| Posted: Tue Jan 16th, 2024 20:48 |
|
177th Post |
Iain
|
Good photo Eric. The waters get deeper and deeper with the AI thing.
|
||||||||
|
| |||||||||
| Posted: Wed Jan 17th, 2024 15:08 |
|
178th Post |
Eric
|
Iain wrote:Good photo Eric. The waters get deeper and deeper with the AI thing. Well, I think I’ve done enough trialling on old photos now and have come to the conclusion that it can be a very useful feature if you want 1) to seamlessly and quickly remove unwanted details from an image 2) to do a reasonable job of adding scenes/settings/backgrounds to improve the captured image. It’s by no means perfect but in time it will improve. The issue will always be, why are you doing it? Commercial graphic design activity (where allowed) will no doubt find it incredibly useful and for them the ethics of it all will be important. For us retired togs or hobbyists I can see it being an infrequent distraction, mainly because it’s only us as individuals that care what the final result looks like. With that I think I’ve said enough on the subject.
____________________ Eric |
||||||||
|
| |||||||||
| Posted: Thu Jan 18th, 2024 14:38 |
|
179th Post |
Eric
|
I said I would shut up but I have discovered one further 'snag' that's worth mentioning.....one last post, honest. As previously mentioned, replacing backgrounds was a frequent event while creating brochures back in the day. It was often necessary to give extra space around the product for text boxes/ slogans/logos etc. However leaving white space wasn't sexy enough so we often added a complimentary background. This photo of a custom built car was a rush job. The car (the only complete fully built model they had in the workshop) was being collected by the purchaser the next morning, so we only had the afternoon to photograph it ...and it couldn't be driven anywhere !!!!! So the back yard of the garage was the only location....and yet they wanted lots of location scenery shots with the car for the brochure. This was before digital cameras. Medium format negative scanned, manually masked and the background ( a Lake District holiday photo courtesy of my wife I believe) was added to this one.  Click here to comment on this image. So today for fun, I decided to replace the background using smart Object masking and Generative fill with a phrase "Lake district landscape" The smart object masking wasn't smart enough to include the 'see through' parts of the image....not surprising....It missed most of the quarter light, an area around the head rest and a see through the windscreen next to the pillar ...also on the other side (not shown in this close up). I therefore manually included these areas in the mask but excluded the road.  Click here to comment on this image. It generated an adequate image but the fill didn't include the areas I had manually added to the mask.  Click here to comment on this image. The strange thing is, all areas were included in the mask as you can see by the mask layer below....  Click here to comment on this image. AI decided not to include them!! As well as excluding manually added masked areas, AI also did some strange edge effects. It removed / trimmed areas around the front and back of the car as well as trimming the roof and 'fudging' the head rest area. (red dotted outlines compared to original photo)  Click here to comment on this image. Of course it will always be hard for ai to interpret the effect of multiple layers of glass and 'tone down' the background image in the areas were they occur. I have also noted that in small discrete isolated areas of a mask (like around the headrest in this image) AI seems to get confused trying to squeeze in detail that's a bit unreal. As mentioned previously, one could over lay the original cut out to recover the outline more faithfully....but not so the see through parts within the main outline. My final take from this is this is a LONG way from hands free photo editing. It's a bit like getting a child to paint a picture in the lounge. You need to make sure you've covered everything you don't want messed up and still stand over them to prevent random flicks and creative correction to the sofa or coffee table. Graphic designers and photo manipulators are safe for a while.  I would just add that you can edit the smart mask selection criteria, so maybe this performance could be improved ......by ME or YOU adjusting the settings manually.
____________________ Eric |
||||||||
|
| |||||||||
| Posted: Thu Jan 18th, 2024 15:30 |
|
180th Post |
Iain
|
As you say its not perfect, YET, but I think it might get there.
|
||||||||
|
| |||||||||
| This is topic ID = 1861 Current time is 19:36 | Page: ... 13 14 15 16 17 18 19 20 21 22 23 ... | |
| Nikon DSLR Forums > The Image Processing and Editing Forums > Software for Image Processing > Using AI in photography | Top | |
Users viewing this topic |
||
Current theme is Modern editor
| A small amount of member data is captured and held in an attempt to reduce spammers and to manage users. This site also uses cookies to ensure ease of use. In order to comply with new DPR regulations you are required to agree/disagree with this process. If you do not agree then please email the Admins using info@nikondslr.uk Thank you. |
Hosted by Octarine Services
UltraBB 1.173 Copyright © 2008-2025 Data 1 Systems
Page processed in 0.0833 seconds (73% database + 27% PHP). 83 queries executed.
UltraBB 1.173 Copyright © 2008-2025 Data 1 Systems
Page processed in 0.0833 seconds (73% database + 27% PHP). 83 queries executed.