Sushi, a classic representative of Japanese cuisine, is a delight enjoyed by many
Because of this, the demand for photography is quite high, and restaurants frequently request my services to take photos.
As technology advances, it’s becoming more common for people to use AI-generated images instead of taking photos themselves. Honestly, this is impacting my work (heading towards the unemployment line for photographers +1).
To sustain my livelihood, facing reality and carefully examining the differences between the sushi photos I take myself and those generated by AI is essential. Let’s explore together!
Here are two photos for comparison:
Without revealing the answer, which one do you think is AI-generated?
The answer is the one with four pieces of sushi.
First, I must say that as of early 2025 (when I’m writing this), AI-generated photos have reached a “usable” level and are incredibly convenient. It takes just a few seconds to produce (4 photos, different styles, each time!!) at the push of a button. If used in a menu with small images, they won’t pixelate and are indeed sufficient (get ready for unemployment!!).
However, if we look more closely, there are still several issues with AI-generated photos that cannot be resolved:
Firstly, consistency
With AI generation, while you can “limit” certain aspects with commands, you still can’t control the output 100%—achieving completely consistent lighting, angles, precise positions, ingredients, etc., is currently not possible. For single-photo designs, this isn’t a problem; however, for menu designs, where a series of photos need to be consistent in style, the lack of uniformity can make the final product feel less cohesive.
Using my own photos as an example, I can achieve a “series” with a uniform feel, without any discord. This is indeed a significant difference.
Secondly, authenticity and uncertainty.
With relevant legal regulations, such as the Trade Descriptions Ordinance in Hong Kong, using AI-generated images to display food can expose you to challenges from customers and potential legal issues; additionally, AI’s uncertainty means you can’t control the generated food’s portion size, number, amount of ingredients, etc. If the actual dish differs from the AI image, problems will arise…
Thirdly, it’s always about aesthetics.
While AI food photos are beautiful, with an almost painting-like finish, they lack the ability to finely control the aesthetic presentation of the food itself. For example, if my client or their sushi chef insists that tuna sushi must be cut a certain way to be considered beautiful, no matter how many times I generate images using AI (I tried 10 different prompts, generated 40 photos, still didn’t meet the requirement).
When you or your client have high and specific standards, AI definitely cannot be relied upon.
Fourthly, I finally started carefully comparing the sushi photos I took myself with AI-generated sushi photos to see the differences:
- Meat texture: AI has flaws—some images show mismatched meat textures or fat patterns between the side and surface, which can be noticeable; in contrast, cameras can capture the minute details of sushi, such as the fish’s texture and the graininess of the rice.
- Lighting and shadows: AI mimics well, sometimes even better than my photos. However, it can’t create a specific single light source to set the mood.
- Rice grains: After comparing several photos, I chose this AI photo for comparison because its sushi rice grains were the best without issues. However, sometimes there can be oddities like strange-shaped rice grains.
Other differences… not too significant.
Conclusion: If you don’t have special requirements, AI-generated sushi photos can suffice.
However, if you have high standards and want to create a great menu, sincerely, AI photos are currently unable to assist you, at most serving as a supplementary element (hopefully, I won’t lose my job because of this…).