How We Saved a Client Hundreds of Hours: Machine Learning for Automatic Background Removal on Thousands of Product Photos

The client had tens of thousands of products and needed the background removed from some of them. Our first attempt with a manual checkbox? Purgatory. The solution? We trained our own ML model that automatically recognizes which photos need their background removed. And it works. We had zero experience with machine learning, but a promise to deliver the feature we needed changed that.

The Problem: Ten Thousand Checkboxes
That Nobody Checks

When the client came to us asking for automatic background removal on product photos, it sounded simple. Then reality hit — they have ten thousand products. And not all of them need the background removed.

The first iteration was a classic: we added a checkbox to the admin. The client ticks it for each product to decide whether they want the background removed. A technically clean solution, right?

It worked for a while. And it would have kept working. So where's the problem?

Imagine having 10,000 checkboxes in front of you. How many would you tick before you lose it? We'd guess about 47.

We needed automation. And then it hit us: What if we let the computer decide?

The Solution: A Machine Learning Model
We Trained Ourselves

None of us had any experience with ML. Zero. But we had a problem that needed solving. And most importantly — we have Google and chat AI apps.

Teachable Machine

We came across Teachable Machine from Google. A simple tool where you train your own model right in the browser. No code, no complicated frameworks.

Here's how we did it:

1. We collected training data — a few dozen product photos where the background should disappear.
2. We added a counterweight — photos where the background must stay (like products shown in context, or lifestyle shots).
3. We trained the model — Teachable Machine worked its magic, and within a few minutes we had a working classifier.
4. We exported and deployed it — we downloaded the model and got it running on our server.

How It Works in Production

When a new product photo is uploaded to the e-shop:

1. The model checks it instantly — "Should this photo's background disappear?"
2. If yes → it's automatically sent to the background-removal service.
3. If no → it stays as it is.

The client doesn't have to click any checkboxes. The photos process themselves. We saved them hundreds of hours of manual work.

What We Learned
(and What We Can Do for You Too)

At the start of the project, we had no idea about machine learning or hosting models. But within a few hours, we:

• ✅ Trained our own ML model
• ✅ Deployed it to production
• ✅ Automated a process that would otherwise have taken months of manual work

And now we know we can repeat it. For anything.

Where Else Can ML Help?

• Photo Quality Detection — automatically flag blurry or poorly lit product photos.
• Product Categorization — the model recognizes which category a product belongs to just from the photo.
• Compliance Checking — find out whether product photos meet your internal standards.

Machine learning isn't sci-fi. It's a tool that can save you time, money, and nerves.

Need to Automate
Something Similar?

Do you have thousands of products and manual work is slowing you down? Or do you keep doing the same thing over and over and wonder if it could be automated somehow?

Let's talk about it. Even if it sounds complicated, there's often a simpler solution than you'd think. And if we pulled it off with ML, we can pull it off with your project too.

Write to us or just give us a call. We'd love to show you how automation can transform your business.

P.S.: Teachable Machine is a genuinely great tool if you want to get started with ML without diving into the TensorFlow documentation. Give it a try — you might be surprised by what you can pull off too. 🚀