Mike Lewis
Staff Member
So I want to get it out there RIGHT now that I am presenting an image here that has been AI enhanced. It is NOT something I would really ever do, except as a test. This originally came about because of an original non-AI-enhanced image I posted in the Salvage Week thread, where I commented that I had kinda cut off some of the elements, particularly on the right side. In the feedback someone mentioned AI, and it got me to thinking. So I decided just as a fun exercise to use Photoshop generative Fill to 'add in' some missing pieces. Now, given the complexity of the foreground in particular I expected the AI result to look ugly and be an obviously manipulated image. So I extended the canvas on both sides, drew some partially overlapping selection boxes on each side one at a time, and then invoked the AI based Generative Fill command with the prompt to simply 'extend the existing image'. I was honestly blown away by the results.
Can you tell it has been modified in this way? Well, at 100% you can see what seem to me to be pretty subtle flaws. The overall image gets softer where the AI content has been added, there are some very faint stitching details, and probably something else if I cared to REALLY look. But I also feel that if I had posted this AI version as the original image here on FW (especially at 3000 pixels max width) no one would have batted an eye - it seems to be that good.
I find this result both amazing and a little dismaying as well. I mean, we must be very close to the point where images like this can be created from scratch and are going to be indistinguishable from true photographs of the physical world. I of course knew we were heading there, but I did not think we were this close. While I do not actually see anything wrong with this type of an edit as long as it is revealed to the viewer what was done, I do not plan to utilize these kind of edits in my own work. I guess one could argue that using content aware fill to remove a rock or a branch (something that I have done on occasion) is a similar thing, but to me it doesn't feel the same. Removing some small detail that was there feels less intrusive than creating entirely new content that is added to the image. Perhaps I am kidding myself on the magnitude of that distinction. But the debate does not matter really, as this technology is going to keep being enhanced as time goes on and before we know it the output from these tools will be capable as passing for the output from the best camera gear out there. And there will be plenty of folks happy to exploit that capability, often without full disclosure. That seems like a scary future to me, for many reasons. What do you think?
Here is the original image, with no extension of the edges of the frame, just the way it came out of the camera as far as what is visible is concerned. Already posted elsewhere but reposted here for comparison:
Here is the AI enhanced version. At the resolution allowed here on FW I think it is very hard to see it has been enhanced. You be the judge...
ML
Can you tell it has been modified in this way? Well, at 100% you can see what seem to me to be pretty subtle flaws. The overall image gets softer where the AI content has been added, there are some very faint stitching details, and probably something else if I cared to REALLY look. But I also feel that if I had posted this AI version as the original image here on FW (especially at 3000 pixels max width) no one would have batted an eye - it seems to be that good.
I find this result both amazing and a little dismaying as well. I mean, we must be very close to the point where images like this can be created from scratch and are going to be indistinguishable from true photographs of the physical world. I of course knew we were heading there, but I did not think we were this close. While I do not actually see anything wrong with this type of an edit as long as it is revealed to the viewer what was done, I do not plan to utilize these kind of edits in my own work. I guess one could argue that using content aware fill to remove a rock or a branch (something that I have done on occasion) is a similar thing, but to me it doesn't feel the same. Removing some small detail that was there feels less intrusive than creating entirely new content that is added to the image. Perhaps I am kidding myself on the magnitude of that distinction. But the debate does not matter really, as this technology is going to keep being enhanced as time goes on and before we know it the output from these tools will be capable as passing for the output from the best camera gear out there. And there will be plenty of folks happy to exploit that capability, often without full disclosure. That seems like a scary future to me, for many reasons. What do you think?
Here is the original image, with no extension of the edges of the frame, just the way it came out of the camera as far as what is visible is concerned. Already posted elsewhere but reposted here for comparison:
Here is the AI enhanced version. At the resolution allowed here on FW I think it is very hard to see it has been enhanced. You be the judge...
ML
