The Covid-19 pandemic brought about an unprecedented cycle of digitally spread humor. This article analyzes a corpus of 12,337 humor items from 80+ countries, mainly in visual format, and mostly memes, collected during the first half of 2020, to understand the features and intended audiences of this “pandemic humor”. Employing visual machine-learning techniques and additional qualitative analysis, we ask which actors and which templates were most prominent in the pandemic humor, and how these actors and templates vary on the following dimensions: local vs. global, Covid-specific vs. general, and specifically for the actors, political vs. not political. Our analysis shows that most pandemic memes from the first wave are not political. The vast majority of the memes are global: They are based on well-recognized meme templates, and almost all identified actors are part of a cast of set “meme faces”, mostly from the US and the UK but recognized around the world. Most popular templates were found in several countries and languages, including non-European languages. Most memes were based on non-Covid specific templates, but we found new Covid-specific memes, which sheds new light on the process by which memes emerge, spread, and potentially become new meme templates. Our analysis supplements existing studies of (Covid) memes that mostly focus on small national samples, using qualitative methods. This cross-national analysis is enabled by a global dataset with unique data on geographical origin of humor. We show the usefulness of visual machine learning for identifying the emergence, spread and prevalence of transnational (humorous) cultural forms. By combining large-scale computational analysis with in-depth analysis, we bridge a gap in in meme studies between (mostly quantitative) data sciences and (mostly qualitative) communication and media studies.