At Ubilabs, it is a tradition to create technologically innovative Christmas cards every year. This year, for many, the focus was on AI. That's why we came up with something special: A symbiosis of analog postcard mailing and technological progress, with a special focus on the topic of artificial intelligence. The result: personalized Christmas cards for hundreds of customers that would not have been possible without the support of AI.
Image and text generation: A personal touch thanks to AI
AI enables many things that previously simply took too much time. This includes the creation of individual, personalized cards for each of our customers. No design should be duplicated, and each contact receives a unique message in the form of a haiku to match the company. We used two key elements of AI to make each card unique: image and text generation.
Image generation with Stable Diffusion
To generate the images, we used the open source software Stable Diffusion, which runs on the Google Cloud. We generated hundreds of winter landscapes in which our trademark is embedded as a building. The ControlNet Canny Model was used for this, which provides a fixed framework for image generation. With the corresponding prompt, we created around 500 unique A6 card designs.
Text generation: Personalized Haikus
The trick was in the personalization: Using Large Language Models, we generated three to four keywords per company based on the language of the contact and the company name. We then used OpenAI and Vertex AI to create seasonal, Christmas-themed haikus from these keywords. This process was automated via a Python script in Google Colaboratory, which processed the data from a Google spreadsheet and inserted the haikus directly.
AI foundation enhanced with human touches
The results served as a baseline, but some human tweaking was still needed. We checked everything manually and carried out several rounds of personal improvements to match the haiku content to the companies and maintain a festive character. This resulted in hundreds of unique cards, which we converted into print data using additional scripts.
We wish everyone a Merry Christmas and look forward to the new technical developments that the coming year will bring!