Cellpic Sunday – Experiment with Generated Fill

Lake Pontchartrain Causeway.

Metairie, Louisiana.

I’m not a regular Adobe Photoshop user, even though I have a Photography subscription plan. I find Photoshop to be more work than it’s worth to me so unless I am doing something really special, I use Adobe Lightroom Classic mostly.

I have been intrigued by the Generative Fill AI tool in Photoshop as I’ve seen several videos of the Beta version in action. I vowed not to mess with beta versions of software many years ago, and I stuck to it. Well, recently Adobe released Photoshop 2024, and Generative Fill is now included in the program. Crossing the Lake Pontchartrain Causeway Bridge presented an interesting example of how I could use the new AI tool to improve an image.

I was sitting in the back seat behind the driver in a Chevy Suburban. Since there are no “rest areas” or “scenic views” allowed on the causeway, I could only capture images from within the moving vehicle as we traveled at 65 MPH across the bridge. You can check my recent Travel Tuesday post here to see the images I captured as we crossed the largest overwater bridge in the world.

About the photo: I used my Samsung S23U to capture the panoramic view above, but it was marred by the presence of the windshield frame and rearview mirror. Here is the original image.

Original photograph from the Samsung S23U.

The first thing I did was send the image to Photoshop for my first published experiments with Generative Fill. The newly installed program launched with some legalese prompts regarding the new AI tool. Once the image loaded, I immediately used the lasso tool to draw a rough outline around the rearview mirror assembly. From the videos I’d watched, I knew to be sure to include some of the sky in my selection so Photoshop’s AI would know what to replace the selection with.

Image with the rearview mirror removed.

The tool generated three options and I picked the best of the three. All three included a section of the dark blue window tint at the top of the windshield. My second generative fill application included a selection of those darker blue areas. I used the text box to indicate “remove color cast” and the program did a pretty good job of removing the blue tint. I left the small section of blue tint that remained in the final image for you to see how well it did. I could have easily done some manual editing to completely eliminate the darker blue hue in the extreme upper right of the final image.

Finally, I used the lasso tool to make a rough circle around the dashboard, steering wheel, and the A-pillars on each side of the windshield. I was amazed at how well the AI extended the roadway and sidewalls into the bottom of the image. One of the three choices did have some distortion, and if I didn’t like any of them, I could have created three more tries with a single mouse click.

From there, I sent the image back into Lightroom Classic to correct the tilted horizon and then touched up the photo with my normal processing techniques. I can see Generative Fill will be useful to me on special occasions, and I can also see its potential for creative photography. As is my usual policy, whenever I use a tool like Generative Fill or replace a sky, I’ll be sure to ‘fess up.

I encourage fellow bloggers to create their own Cellpic Sunday posts. I never have a specific topic for this feature, and the only rules are that the photo must be captured with a cell phone, iPad, or another mobile device… If you have an image from a drone or even a dashcam, that’s acceptable as well. The second rule is to link your challenge response to this post or leave a comment here with a link to your post in the comment. Oh, and, you don’t have to post it on a Sunday.

John Steiner

51 comments

  1. You’re having a good time with Photoshop and with great results. In my experience, if not with the same end result in mind, has been more of a mixed bag. Removing small things work fine but don’t ask it to do much from scratch… the results can be very funny, tho.
    Here is mine:

    Cellpic Sunday – Glasgow

  2. A really interesting walk-through of this process. It has done a very good job, especially in the lower part of the image. It would take ages to create this manually, if indeed it were possible!

    • I might have tried to remove the mirror by hand, not sure how that would work. I’d end up cropping out the bottom to eliminate the dashboard.
      It’s uncanny how the AI seems to know what to do (most of the time.)

    • When I started playing with the tool myself, I began to see more and more places it could be useful. I have many images spoiled by some small thing (like a photobomb) that I can probably revisit now. 🙂

  3. Wow, John, your before and after images of the process for capturing this lovely image are so interesting and quite different! Very nicely done. That is quite the span to drive over but photo magic and smart photography really sells the shot! Here is my post this week with a few cell pic shots 🙂 https://secondwindleisure.com/2023/11/12/sunday-stills-monthly-color-challenge-all-the-leaves-are-brown-and-much-more/
    BTW, I still have a lot to learn with the pro feature on the S23U. I had a chance to see the northern lights from our house again, and I was pleased I could capture them a bit with the phone. More about that later.

    • I look forward to your northern lights photos. I’ve only seen them a couple of times and never when I had a camera ready.
      I will have lots more examples of Generated Fill and the new AI tools being added to Luminar Neo.
      I will always note when I substantially modify an image, (except maybe for removing the errant sign or misplaced distracting object.)

  4. I also take a ton of pictures during a journey and many need to be immediately binned. But I still have a tob because I see potential in them with the hope of AI helping me out 😀. I’m not on photoshop because it feels very cumbersome to me too. Hoping that other more user-friendly tools will also become better soon 😉.
    My entry here:

    Icon

    • I have already experimented with a few older images to some success, but at this point, I don’t have a reason to share them on the blog. Maybe I’ll gather a gallery together with a specific purpose to feature.

    • I expect that there will be something come along. Lightroom for the iPad has a tool called Content Aware Remove that has similar, though less sophisticated options. I can’t speak to its efficiency, though. I don’t have it on my iPad, but I have watched a few videos.

  5. Love this post, John. Have been watching the PS ads on Telvision for generative fill and wondered if it would be as good in real life, so to speak. Your result is outstanding. Many thanks.

  6. This detailed exploration of the Generative Fill AI tool in Photoshop showcases its potential for enhancing images, especially in challenging situations like capturing photos from a moving vehicle. The step-by-step process, accompanied by before-and-after visuals, provides valuable insights into the tool’s capabilities. The author’s transparency about the tool’s impact on the image, along with the commitment to ‘fess up’ when using such features, adds authenticity to the discussion. The article not only serves as a tutorial but also inspires fellow bloggers to experiment with cell phone photography and digital tools for creative expression.

      • You’re welcome! I’m glad to hear you find the tool fascinating. Sharing experiences adds a valuable perspective. If you have any specific use cases or insights you’d like to share, feel free to let me know. Your input is greatly appreciated! 🌟

Leave a comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.