A Week of Grey Skies in Rotterdam: Three AI Sky Replacement Tools Tested
Looking back through the 47 photos I took at the Port of Rotterdam last week, I let out a sigh. Every single sky is the same. Dense grey, slightly lighter grey, and somewhat darker grey. The sky stretching over the Erasmus Bridge, the backdrop behind the cranes on the Katendrecht side, all the same tone. That's just October in the Netherlands, I know. But not having a single shot worth putting in my portfolio is a different matter entirely.
So I decided to test three AI sky replacement tools that have been getting a lot of buzz, applying them to the same set of photos. I split the test into three categories: shots with complex overlapping port cranes, shots with water surface reflections, and shots with simple building outlines.
Luminar Neo's Sky AI
I pulled out Luminar Neo first. It's the tool I've used the longest, and I had decent expectations given its fairly large sky library. The results were mixed. On shots with clean building outlines, the replacement blended in surprisingly well. There's a feature that automatically adjusts foreground tones to match the new sky colours, which cut down the disconnect between the dull originals and the fresh sky considerably.
The problem was the crane photos. The AI couldn't handle the fine gaps where the original sky showed through the lattice-like metal structures. Patches of grey remained between the crane framework, and with the newly applied orange sunset wrapping around them, the compositing was obvious. You can manually refine the mask, but after spending more than 15 minutes on a single crane shot, you start thinking, "Why am I even doing this?"

Photoroom vs. Photoshop Generative Fill
Photoroom is primarily known for product photography, but I included it after hearing its landscape editing had recently improved. Being web-based and requiring no separate installation was a plus. It was fast, too, roughly 8 seconds per image.
But the results were too "clean." They looked like renders, not photographs. A perfectly smooth sky placed over the rough textures of the port (rusted bollards, oil-slicked water) made the entire photo feel fake. Could I use these in a portfolio? Honestly, I'd lose credibility the moment I sent them to a client.
The third tool I tested wasn't a standalone app but Photoshop's Generative Fill feature. You roughly select the sky area and enter a prompt, and this is where I got unexpected results. Typing "dramatic sunset" produced over-the-top output, but being specific with something like "overcast sky with slight warm break in clouds" generated quite usable skies. The handling of fine gaps between cranes was also the best of the three tools. Not perfect, but it noticeably cut the time needed for manual touch-ups.
The downside of Photoshop's approach is inconsistency. Even with the same prompt, you get a different sky every time. When you need to match a series of shots to one consistent tone, Luminar Neo was actually more efficient since it can apply the same preset sky repeatedly.

Did Any Actually Make It into the Portfolio?
After running all three tools, the number of photos that actually made it in was 4 out of 47. Two building outline shots processed with Photoshop Generative Fill, one water reflection shot processed with Luminar Neo, and one black-and-white conversion uploaded as is from the original.
Not a single crane photo made the cut. No matter which tool I used, the boundary between the complex metal structures and the new sky never looked natural. That's a clear limitation of current AI sky replacement tech. It works almost flawlessly on simple skylines, but the moment the structure gets complex, the compositing shows.
In terms of cost, Luminar Neo's annual subscription is roughly €89, Photoroom's Pro plan works out to about the same on a monthly basis, and Photoshop is included in the Adobe Photography Plan, so there's no extra cost if you're already subscribed. If you already have Adobe, Generative Fill alone is more than enough to get started.
The week of grey skies wasn't a total waste. At the very least, I now have a clear picture of how these three tools actually perform, and next time it's overcast, I'll probably compose my shots differently during the shoot itself. Keep the skyline simple; frame the cranes as silhouettes. Those 47 grey-sky photos taught me that setting up compositions the AI can handle cleanly is what saves you two hours in post.
Comments