GenAI vs. Photoshop: Why it isn’t quite there yet.

11

October

2024

No ratings yet.

(Gen)AI is a term which is being mentioned endlessly in improving real world workflows. So I thought to myself let’s see if I can use this to my advantage to photoshop a nice picture I have; but which isn’t quite there yet. Mind you, I didn’t have any Photoshop skills prior to this endeavor.

I had a nice picture which I wanted the background to be cleared up. I have seen many social media posts where they made this work flawlessly. So I went into this with the idea that it would be easy, and to a certain extent it was. Take a brush and highlight what you need replaced.

However the AI seems to take the liberty of making changes you never asked for. The picture at hand can be found below. The request was simple; clear the background and replace it with the building behind. The blurry image almost suggest that the AI isn’t able to recognize a simple building pattern with the windows.

This picture was by far the best rendition of me trying out this tool. A major hurdle to get over was that the AI wanted to head into a different direction than the prompt given and area selected. You cannot see it that well in the above picture but its already ‘eating away’ at the front tire which isn’t circular anymore. With this final Product I might be able to upscale the picture myself, but that wasn’t the expectation when I started this little project. After having wasted a few hours I laid the work done and more or less came to the conclusion that a professional would have been able to complete the project Quicker and deliver a higher quality product.

To at least give the AI some credit, the color grading and mixing is pretty good, especially since the area I tried to edit is quite large at a high resolution. And the picture could have come out quite realistic, if it wasn’t for the bad pattern recognition that the AI is trained on.
Moreover, with ‘realistic’ AI generated images I find the pictures to be quite soulless, as the lighting seldom gets captured in any good way.

Finally, I do think that it is inevitable that with more time, resources and more complex learning models AI will be able to capture more photorealism. But even once this is achieved, I don’t believe Photographers need to worry about their job if they are able to implement tools like these themselves.

Please rate this

Leave a Reply

Your email address will not be published. Required fields are marked *