Photoshop and AI continued.

Adobe Photoshop Beta Boggles More

I threatened to dig deeper into Photoshop Beta Generative Fill.

My portrait of my trusty A6400 with a favorite general-purpose varifocal — Sony E PZ 18–105mm f/4.0 G OSS (27–158 equivalent).

From my previous story — Adobe Photoshop Beta’s generative fill is astonishing, and I have not even tried typing in a prompt. To learn it I need to go further than using it as a glorified content-aware fill. When I do, I’ll write a follow-up on my progress.

But this is not about the camera. It is about the Photoshop Beta Generative Fill that I prompted to fill the original neutral background with “Deep space with starfields, suns, and planets.”

My starting point, a simple product shot of a favorite camera/lens combination— All photos herein ©Charles G. Haacker, Author.

I’ve been reading blogs and publications and watching many tutorials on using this marvelous new toy best. It isn’t perfect yet; that’s why it is in Beta, but its path is clearly upward. If we think it is good now…

After making the original picture, I took it into Photoshop Beta and used the pen tool (most accurate with clean lines and curves) to cut the camera out of the background, removing the ball head support and the strap that I am too lazy to unbend from the grommets because I can never remember how to put it back.

Three variations on the theme. In the beta version, you can ask for as many renders as you like, but that may not be true when it makes it to regular Photoshop.

Once I had the cutout masked, I selected everything around it, activated Generational Fill, but this time, instead of leaving it to its own devices, I typed in a prompt, something on the order of deep space starfields with suns and planets.

The tool thinks about it and spits out three choices, as you see. I liked all of them, so there they are.

For comparison, below is an example of making only a rough lasso selection and giving Generational Fill its head. I felt the paths were distracting, and the sign needed to go. This is likely to remain my principal use of this marvelous tool. The original snapshot is transformed into a photoillustration.

These sculptures of mantids are in Lauritzen Gardens, Omaha, Nebraska. — ©Charles Haacker.

Notice how, in the picture on the left, the AI has dropped a realistic shadow on the massif.

Returning to the camera, I typed a prompt asking for mountains and evergreens. These were what it spit out. I was surprised when it opted to have a woman’s hand supporting the camera. And it’s not very good. Her skin is mottled, her thumbnail is ragged, and she’s either been scratching her mosquito bites or bitten by a vampire. We all know that AI doesn't do well with hands in its present state of the art.

The last prompt I typed was “urban landscape.” I was pleasantly surprised that the AI chose mostly monochrome backdrops that went well with the camera.

I was impressed that the AI used a sheen of water, making for both reflections and shadows.

This screenshot may give you an idea of the process, which with practice, gets easier. I may write another piece if I have more samples to show.

There will be issues, ethical questions, and concerns about Photoshop encouraging deepfakes. I have been gaga over what I can do with modern photo editing technology that gets better almost daily. But I am honest and ethical. I try always to use the term “photoillustration” when I’ve done something I consider significant. On the other hand, if I clone out a distracting fireplug, I may not think it's worth the mention. I’m neither a photojournalist nor a forensic/evidence photographer. I’m just a kid havin’ fun.

The bad guys, though, can do things once deemed impossible and do them so seamlessly as to be nearly indetectable. Are there, can there be good solutions?

I dunno. I got nuthin’.

📸As always, gratitude for looking in. I sincerely appreciate it! 😊👍