AI product photography: I rebuilt a shoot that cost me $1,300

In 2020 a Saudi client sent me a brief: skincare products, mostly argan oil based, and they needed a shoot for social media.

I did the work. It took thirty days and cost me $1,300 across studio, lighting, time and production. That was roughly sixty percent of what the client paid.

This month I went back to those same photographs and tried to produce them with AI.

The numbers#

I used Seedream and Nano Banana. The result shocked me, and I am a photographer.

Four hours instead of thirty days. Fifty dollars instead of thirteen hundred.

And my honest first thought was: I wish I had had this then. I would not have lost sixty percent of my fee.

What genuinely worked#

Backgrounds and moods. The same product on marble, on wood, in morning light, in the shadow of a window. Each of those used to mean rebuilding the set.

Variations. Twenty crops at different sizes for different platforms. Entirely tedious work.

Impossible scenes. The product among argan fields, which used to mean travelling or buying a stock image that matched nothing.

Lighting. Seeing the same bottle under five lighting setups before committing, instead of committing and then discovering.

What did not work#

And I have to be precise here, because most articles stop at the part above.

The actual product. If a client sells one specific bottle, with its label, its Arabic text and the shape of its cap, generating from scratch gives you a bottle that resembles theirs. The client sees the difference instantly.

The approach that worked: start from a real photograph of the product and change what surrounds it. Do not generate the product.

Arabic text on packaging. It comes out mangled in most attempts. A missing letter, a reversed word, or a shape that looks like Arabic and does not read.

Consistency. Five images of the same product come back with five small differences in size or colour. The eye catches it in an Instagram grid even when it cannot name it.

Hands. Still a problem. Any scene with a hand holding the product needs many attempts.

The method I now recommend#

Photograph the product once, properly. White background, clean light, several angles. That single hour is the foundation for everything after it.

Generate everything around it. Backgrounds, moods, seasons, occasions.

Check the Arabic text letter by letter. In every image.

Keep the original. When the client asks "is this my product?", the answer is yes, and that layer is real.

What this means for you#

If you shoot: the part that was billed by the hour collapsed. What remains is the first hour, photographing the product correctly, and the judgement about what looks good.

If you run a small shop: photograph your products on a white background with a phone, then generate the rest. That puts you close to large retailers' imagery at close to zero cost.

If you run an agency: reprice this service before your client does it for you.

In closing#

The same job: thirty days and $1,300, against four hours and $50. That is not a prediction about the future. It is a comparison between something I did myself, twice.

Try this today: photograph one product on a white background with your phone and generate five different backgrounds for it. One hour will show you where the line now sits.

Common questions#

Does AI replace product photography?

It does not replace photographing the product itself, only everything around it. The method that works is shooting the product once, properly, then generating backgrounds, moods and variations.

Why not generate the product itself?

Because the result resembles your product without being it, and clients see the difference immediately in the label, the text and the cap. Start from a real photograph and change the surroundings.

What is the biggest problem with generated images for Arabic products?

Arabic text on packaging comes out mangled in most attempts, consistency across five images of the same product, and hands holding the product.

How much does this actually save?

In a test rebuilding a real 2020 shoot: thirty working days became four hours, and $1,300 became $50 in tokens.