Ps beta - Generative fill

[MFloyd]

Here at the 24H of Le Mans. Over 300’000 attendance; 1300 accredited photographers. Very difficult to have an ideal point to shoot from, together with some undisciplined colleagues .


version edited with Ps beta

The original picture was cluttered with the head of a photographer, scaffolding, and a hat.


original version

Things went even worse when at the start a “pro” photographer suddenly raised his camera above his head ruining all my pictures together with the ones of dozens of my colleagues.

_D654314.jpg by Christian Vermeulen, sur Flickr
                             
By chance the generative fill from Ps beta saved the game.

_D654314-3.jpg by Christian Vermeulen, sur Flickr

[Bent Hjarbo]

The function works fantastic, but will it still be photojournalism?
Maybe the problem is smaller with the things you photograph.

[Thomas Stellwag]

 is the "pro" still alive?

it seems the the fill function takes out saturation, nevertheless the results are impressive
for me the most obvious point is the car behind the camera, that looses one headlight by filling

[MFloyd]

The function works fantastic, but will it still be photojournalism?
Maybe the problem is smaller with the things you photograph.

Genuine question. Therefore I footnoted the photographs that some parts of the image were AI filled.

is the "pro" still alive?

it seems the the fill function takes out saturation, nevertheless the results are impressive
for me the most obvious point is the car behind the camera, that looses one headlight by filling

Tabards are wearing an identification number. This is one of the reasons: allowing to report. The 24H of Le Mans credentials are very selective, and you can be out very quickly, with no possibility for appeal.

The departure of 56 cars generated a lot of dust, henceforth reduced the saturation. Images were further processed to enhance clarity and saturation inter alia .

Filling is still not perfect, and at this stage, not passing the pixel peeping test. Still a beta version. Nevertheless, it saved me from (total) disaster.

[John Geerts]

Nice comparisions. I am not convinced with the new feature, sometimes is suddenly works, but in many cases not.

Working with the clone stamp (also AI ) is still more effective...

[MFloyd]

I had over 30 pictures or to bin or to edit, in a timeframe of less than 30 minutes. Not sure I could do this with the classical clone tool. Moreover, in most pictures, there was nothing to clone from.


24H Le Mans 2023

[armando_m]

The recent progress of AI has been impressive !
A big problem will be the users and their ethics, not everyone will say this is AI generated

[Michael Erlewine]

The recent progress of AI has been impressive !
A big problem will be the users and their ethics, not everyone will say this is AI generated

I have done more than 10,000 images using Midjourney. I use it not with "Art" in mind, but with illustrating in mind, since I blog to some 11,000 people each day and need something illustrative to have along with the blog.  I always label the illustration with:

[Midjourney graphic prompted by me.]

Prompting, however, is something of an art and is very much creative.

I recently did a blog where I needed an example of an actor that many feel is authentic. Here is what I came up with, Harrison Ford in his role as Indiana Jones, made with Midjourney AI graphics. Of course, I can just ask for an image of Harrison Ford and it will churn one out.

However, to get an image that actually looks authentic takes time and skill using AI. Here is what I came up with. Not bad considering this is not a direct pull from an existing image, but creating an image together from a small universe of scraped data. It took many iterations to get this one.

[MFloyd]

A couple of years ago, I made this pleasant picture of a breakfast table in the south of France.


IMG_0847.jpg

Until I realized that the cup was empty, and therefore I never published. But that was not counting on the “generative fill” AI feature of the last Ps beta version. Therefore, I cercled the upper part of the cup and instructed: “fill the cup with black coffee”

Hereunder the result; there is even a brown creamy border.


IMG_0847.jpg

Off course, Ps provide me with several alternatives; but I took the initial proposal.

[Erik Lund]

That was quite smart The foam and reflection does it!

[MFloyd]

That was quite smart The foam and reflection does it!

Yes. Quite unbelievable. Writing the instruction took me no time; imagine the time needed to achieve a similar result, the traditional way.

[RobOK]

There will be a period of time where the technical prowess blows us away, then it will become common place. Art does not die, it shifts with technology. Image making has shifted from paint to film to digital to AI (simplifying many steps in between). Art comes from the human desire to create and express, and the tools change. The bigger challenge is the inundation of more and more media, streaming past us every moment. I suspect that is why many like this forum for images. Slower, authentic, people creating in their own ways. Technology marches on.

[Akira]

I don't want to sound like a wet blanket.  But the "complemented" parts of both examples look a bit weird.  The left front part of the racing car doesn't look symmetrical to the right front part.  Also, the behavior of the generated people in the second example doesn't fit the context of the original image.

I would believe that these anomalies and weirdness would be improved and addressed, but at least at this moment, this function seems to work mostly for the landscape images where the random stuff like grasses, trees bushes, gravels, etc. can be generated without much contradiction.

[Dogman]

Even with the guy's arms and camera included, I prefer that photo to the one where the whole bottom area is blank and empty.  But I know I'm weird.  I'm just not into perfect pictures.

[MFloyd]

Even with the guy's arms and camera included, I prefer that photo to the one where the whole bottom area is blank and empty.  But I know I'm weird.  I'm just not into perfect pictures.

The blank and empty part is edited out in the print edition. I left it here for comparison purposes. The guy’s arm version is simply not publishable.


_D654314-3.jpg by Christian Vermeulen, sur Flickr