How we used AI for a product photoshoot
Michael Bertel |
We launched a new product (https://limifyze.com/servenia) and we needed pictures for marketing stuff (online and print).
Therefore, we quickly took some photos of the machine.
The photos were OK, but of course they were boring and one-dimensional.
So I finally decided to try out all the fancy image gen AI tools to edit our images.
The day after I had finished creating all the images I needed, Google released Gemini 2.5 Flash Image, which changed my workflow again. If I need new images in two weeks, I am sure that both my workflow and my tools will change again. If so, I plan to update this blog post.
The goal is to create realistic photos of a specific product that already exists in real life, showing it in different environments (i.e. different use cases).
My current workflow (3 September 2025)
1 - First, take pictures of the product in from different angles and distances in a (more or less) clean room.
2 - Gemini 2.5 Flash Image
Tell Gemini 2.5 Flash Image to change the environment.
3 - Upscale
Unfortunately, Gemini 2.5 Flash Image doesn't maintain the quality.
I used upscayl because it's open source but you can use whatever works best for you.
4 - Fooocus
If you are not happy with the image details after the upscaling, use Fooocus to improve them.
My workflow before Gemini 2.5 Flash (before 26 Aug. 2025)
1 - First take pictures of the product in from different angles and distances in a (more or less) clean room.
Remove the background - outpaint, and remove/add stuff with “inpaint” and “remove”
3 - ChatGPT
Use this image to add “models”.
4 - Photoshop / Photopea
"Cut out" the Models with a Mask and merge.
This could obviously be done much better if you have some PS skills.
5 - Fooocus
remove borders
Remove/inpaint anything you don't like.
This worked for me although it’s probably too sophisticated for such a “simple” task.