Features People Profiles
Spotlight: Janis Akmentins, head of engineering at Printful

Printful is a global company that specializes in print-on-demand. We spoke with Janis Akmentins, head of engineering at the company, to discuss how implementing artificial intelligence (AI) and machine learning (ML) in print-on-demand provides customers with better images.

What is Printful trying to achieve by implementing AI solutions?
JA: Nowadays, AI can be used to solve problems in all kinds of areas. We use it for ensuring quality prints and detecting copyright violations. We put a lot of effort into making sure that the products we offer and the designs we print match all the quality standards that we’ve set for ourselves. However, while we have complete control over the quality of the physical products we provide, print quality is harder to ensure due to the design variety.

Our customers are very creative—every day they create numerous designs to put on their products. To keep the print quality in check, design files have to meet certain guidelines. This is where AI steps in.

AI solutions help us inform customers about cases where their designs don’t match print requirements. That way, customers can adjust their designs to get a quality end product.

Additionally, by implementing the Background Removal Tool, we allow our customers to make complex image adjustments straight from the website—they don’t have to use heavy image editing software. This saves our customers time and possibly money as well—if the software they would otherwise use is costly.
Another way we use AI is to detect copyright violations in designs. Our quality assurance team has an eye out for possible copyright violations so these designs don’t continue in the manufacturing flow.

Both copyright violation detection and design guideline matching are internal processes where we have now implemented an AI solution. Before we used AI solutions, all this work had to be done manually, and that affected the fulfillment time. Since we automated these internal processes, we’ve been able to get the orders out faster.

We see great potential for adding more AI solutions in the future.

How does AI work to correct a less-than-perfect printed image?
JA: First, we used image transparency detection. To train the neural network, we first had to create a dataset by manually classifying our customer designs with transparent areas. Based on our knowledge about the printing process, we manually categorized several images on a scale from one to nine, depending on how much impact it would have on the physical print. After creating the dataset, we trained the neural network to automatically categorize the designs.

Next, we tackled image upscaling. In this case, we also had to start with training the neural network. We manually entered two versions of the same image. The first was a low-resolution image and the second one was the same image in higher resolution. During this process, the neural network was able to learn how the pixels differ in these images and what rules had to be applied to reconstruct the image from low to high resolution. Hence, when a customer uploads a low-resolution image, we can upscale it by adding better resolution.

What are some of the biggest opportunities you see in the print industry today?
JA: There are a lot of people whose work is affected by the restrictions set by their governments, and many start looking for additional income sources. The print-on-demand model is a good opportunity because they can experiment with launching their own business with no upfront costs.

Printful is helping our customers do their business. We open new fulfillment centres to be physically closer to end customers and provide quicker shipping.

Our IT team is continuously building tools and coming up with solutions to improve customer experience and make business management more pleasant.

I believe the changes we see now will have a long-lasting impact on e-commerce. I believe we’ll see these trends in the coming years: faster delivery times (with decentralized supply chains), richer user experience (with the help of AI), and an increasing amount of mobile app usage.

Janis’ answers were edited for length and clarity. For more Spotlight Q&A interviews, please visit printaction.com/profiles.

This article originally appeared in the January/February 2021 issue of PrintAction.