We believe that those who think of money as the world’s most important resource are wrong – since the foremost business capital nowadays is time. When choosing products and services, customers are guided not only by the cost criterion but most of all by the desire to achieve a satisfactory result quickly and without unnecessary effort. It is no different in the personalized photo printing industry.
The big idea standing behind our company is to deliver outstanding software allowing to preserve people’s best memories thanks to printed photo products. Nevertheless, we were aware that the traditional approach to generating photo books or albums requires a lot of manual and time-consuming operations.
To meet this challenge, we have developed the Printbox Smart Creation. This time we would like to present the story of our work on this solution, and our questions will be answered by Kasper Rybak, the Product Owner at Printbox.
How the story begins
What was your motivation to begin with the Smart Creation project?
The history of creating photo products with the use of intelligent algorithms dates back to the very beginnings of Printbox. From the first moments of the company’s operation, Michał Czaicki (CEO) predicted that the automation of photo selection and photo book building is the way to success in the industry of personalized photo products. However, to do this, we first had to build and refine all the basic software components.
Therefore, how did your automation work begin?
We started our work on automation with the introduction of more simple mechanisms, such as Autofill and Autocreation. The Autofill feature was a simple way to fill photo slots. In turn, the Autocreation operated on the basis of our first proprietary algorithms that placed photos in the project, taking into account many variables, such as:
- The time the photo was taken and a series of photos to group similar ones on the same pages.
- The amount of photos and pages in the book.
- Available layouts.
However, we wanted to go even further in the development of our tool and adapt it to the needs of the market and consumers. For this reason, we decided to do more in-depth research on the practicalities of user image selection and start working with more advanced technologies.
The development of Printbox Smart Creation culminated in 2020. At that time, we could focus on maturing the advanced automation, since all other elements of our system were already well refined.
What is the Smart Creation by Printbox in practice?
Could you introduce the mechanism of Smart Creation?
We were aware that one of the biggest problems with buying photo books is the very start of the project. People are concerned about having to go through hundreds or thousands of photos to pick the best ones.
That is why we decided to allow users to load as many photos as they want into the editor so that intelligent algorithms would do the tedious work for them. The Smart Creation analyzes each of the uploaded images already at the time of their transfer. The technology considers objective qualities such as:
- are the pictures sharp or blurred?
- whether the exposure is good?
- is there a perceptible noise?
In addition, the algorithms included in Smart Creation analyze the subject of the photo by categorizing it and detect objects related to the topic of a whole picture – their positions and sizes. Based on the info collected, the solution calculates the final score for the image.
Also, during the analysis, the system detects similar pictures, to rule out the duplication of content in the story presented in the photo book.
Rated images are grouped accordingly and organized into consecutive spreads to create a coherent narrative. Grouping is done based on, among others, the dates of picture creation, as well as the visual content of the photos.
During the analysis, the Smart Creation not only reviews user photos but also evaluates the chosen theme. I believe that this is one of the biggest Smart advantages – it doesn’t only create automatically plain white or black books, but it may also automatically create a beautiful project with plenty of clipart, backgrounds, text fields, and other items that give the extra touch to the pages that customers prepared for their end-users.
What kind of technology stands behind the solution?
The editor was created using React / Typescript technology. For photo analysis, we use Artificial Intelligence solutions in TensorFlow with various Convolutional & Siamese Neural Networks paired with classical image processing algorithms. Moreover, we are utilizing cloud solutions, so that we can quickly scale up when required.
Challenges in creating algorithms-based photo products software
What challenges did you face when starting the project and during its implementation?
Our main challenge, which was the driving force for us, was to eliminate all the protracted and cumbersome tasks associated with the process of selecting the best photos, as well as designing and preparing end-products.
However, we asked ourselves if it was possible to automatically create a photo book that the customer deems good enough to buy without batting an eye. This is a process that requires creativity and thousands of images to be searched to select the best ones. There is no “one right way” to create a photo book. Thus, developing algorithms in such an environment is much more difficult.
Moreover, there is no such thing as a “typical” set of photos. Images uploaded by users differ significantly from each other and you should be prepared for even extreme cases.
Could you give an example of such unusual case?
There may be times when a customer uses a set of photos for which 2 photos were dated 1970, 50 images related to an important event organized in 2018, and 100 pictures taken between 2020 and 2021. And it is just the shooting date factor! Add to that the huge dissimilarities in quality, content, as well as differences in styles and themes, and the problem quickly becomes very complex.
What was particularly challenging for you?
I think it was reaching the golden mean between accuracy and speed. Time is money, so all the operations need to be very fast. We need to make sure that each millisecond of processing translates into a visible effect for the final look of a photo book. In the beginning, we set a limit of 65 milliseconds on average devoted to the analysis of a single photo, but we have already achieved this milestone. Therefore, we are working to successively shorten the time needed.
Reception of the Smart Creation among customers
How did customers implement the Smart and what were their reactions?
It is well known that all novelties in business are approached with caution to be able to first identify the potential benefits and risks.
Nevertheless, the initial reception of Smart Creation among our customers was very positive as they understood the capabilities of intelligent technologies.
Since our solution was newly created, we jointly decided to introduce it to the clients’ environment step by step, within single instances, to be able to fully control and analyze all data.
Importantly, along with the first results for groups of end-users buying products in our clients’ stores, we were able to validate our main hypothesis, in which we assumed that we could lessen the time spent on photo book creation and increase the overall conversion.
This confirmed the belief that implementing intelligent algorithms in the photo product sector makes sense and brings real business benefits.
Thanks for the interview, Kasper!
Work on improving and extending the technical capabilities of the Smart Creation continues. A solution of this type requires constant development of the customization options, buying paths optimization, offered themes, and the like.
Our team also conducts a variety of tests, using the results obtained from the customers to introduce data-driven enhancements. One of the examples of recent analyzes are Live Tests carried out from February to May 2021, with the participation of two of our biggest clients.
The results are available in our latest Smart Guide: Intelligent Algorithms in Photo Products Sector. Click here to download the material.