What I Learned Building an AI Product Photography SaaS

Mohamed EL Mansouri — Sat, 29 Aug 2026 13:06:42 +0000

When I started building Shotinger, I thought the hardest part would be the AI.

Getting good images. Finding the right models. Making the generations look realistic.

Turns out, I was wrong.

The AI part is obviously challenging, but getting from “this is technically impressive” to “I would actually use this for my business” is a completely different problem.

After spending time building and testing the product, here are a few things I’ve learned so far.

1. A good-looking image isn't necessarily a good product image

In the beginning, I was obsessed with image quality.

I wanted every generation to look as realistic and polished as possible.

Then I started looking at the problem from the perspective of someone running an ecommerce store.

And I realized something pretty important:

It doesn't matter how beautiful the image is if the product itself isn't accurate.

If you're selling a black hoodie, you don't want an AI-generated hoodie that looks like your hoodie.

You want your actual hoodie.

The logo needs to be right.
The color needs to be right.
The shape needs to be right.
The details need to stay consistent.

That's especially important for fashion and ecommerce because customers are buying the product they see in the image.

So I gradually stopped thinking only about "AI image quality" and started thinking much more about product consistency.

That changed quite a few product decisions.

2. I underestimated how important simplicity is

AI products can become complicated very quickly.

You have models, prompts, styles, settings, references, parameters...

And because all of these things are interesting from a technical perspective, it's tempting to expose them to the user.

But most ecommerce businesses don't care about any of that.

They don't want to learn how image generation works.

They want something much simpler:

Upload a product → choose what you need → get the image.

Making that experience simple is actually harder than it sounds.

You have to decide what the user really needs to control, what can happen automatically, and what should simply disappear behind the interface.

I've learned that sometimes the best feature is the one the user never has to think about.

3. Focusing on a niche makes building easier

There are already a lot of general-purpose AI image generators.

I didn't want to build another one.

Instead, I decided to focus on one specific problem: ecommerce product photography.

That sounds like a limitation, but I've found it to be the opposite.

Once you know who you're building for, a lot of decisions become easier.

You can think about actual use cases:

Product pages
Amazon and other marketplaces
Social media
Paid ads
New product launches
Seasonal campaigns
Fashion and lifestyle images

You're no longer asking:

"What can we do with this AI model?"

You're asking:

"How can we make this particular workflow better?"

That's a much more useful question when you're building a SaaS.

4. The AI model is only one part of the product

This was probably one of my biggest lessons.

It's easy to think that the company with the best AI model will automatically have the best product.

I'm not sure that's true.

The model matters, obviously. But so do:

The user experience
The workflow
Generation speed
Image consistency
Choosing the right outputs
Pricing
How easy it is to get from an upload to a finished image

Two products can use very similar AI technology and still feel completely different.

And I think that's one of the most interesting things about building AI products right now.

The technology is becoming more accessible.

The real differentiation increasingly comes from how you turn that technology into something useful.

5. Building it changes the questions you ask

I'm still figuring a lot of things out with Shotinger.

There are features I'm not sure about. Ideas that looked great on paper but didn't work as well in practice. Things I thought users would care about that turned out to be less important than expected.

That's probably normal when you're building a product.

But the experience has changed the way I think about AI.

At the beginning, my question was:

"What can AI generate?"

Now I find myself asking:

"What problem can I solve with AI so simply that someone would actually want to use it?"

I think that's the more interesting challenge.

AI can generate an incredible number of things now.

Building a useful product around that capability is the hard part.