I asked AI image generators to create the most stereotypical person for every country, running two rounds of the experiment: one in 2023 with Midjourney, and another in 2026 with Grok Imagine, GPT Image, and Gemini. Each run produced separate results for men and women. The creators note these images are AI-generated and do not represent the diversity or complexity of the countries depicted. The project is framed as a lighthearted exploration of how AI perceives different national stereotypes.
In the 2023 round, Midjourney was the model tasked with generating stereotypical figures across countries. By 2026, the comparison expanded to include Grok Imagine, OpenAI's GPT Image, and Google's Gemini. Each model was run separately for male and female outputs, meaning two distinct sets of images per country per model year. The organizers explicitly state the results are not representative of real population diversity—rather, they reflect the biases and learned associations embedded in each model's training data.
Methodology and model comparison
The project invited users to “pick the model at the bottom of the map,” suggesting a side-by-side view of all four models' outputs for any given country. The tagged labels—AI-Generated, Not Representative, and Just for Fun—set expectations upfront. The 2023 run with Midjourney established a baseline, while the 2026 additions allow observation of how later models have shifted or reinforced certain visual patterns. Because each model was asked to produce “the most stereotypical person,” the results are expected to lean into extreme archetypes rather than nuanced portrayals.
Gender separation in outputs
Every country received separate results for a man and a woman. This doubling of the output per model and per region means the experiment captures two distinct stereotypical portraits per nation. The side-by-side presentation likely invites users to compare how each model frames masculinity and femininity differently across cultural contexts. The consistent gender split across all models and years provides a structured data set for informal comparison, even though the results carry the disclaimer that they are not representative.
Disclaimer and critical framing
The project includes three guiding tags. AI-Generated makes clear that every image is synthetic output from a machine-learning model. Not Representative warns that the pictures do not capture the full reality of the countries in question—a necessary caveat given how easily AI outputs can be mistaken for documentary evidence. Just for Fun signals that the exercise is not intended as serious social science or census-level analysis. Together, these tags frame the work as a speculative, playful look at algorithmic perception rather than a factual survey.
What it reveals about AI and culture
By asking each model to output “the most stereotypical person” for every nation, the experiment surfaces how different systems have learned to associate visual traits with national identities. The 2023 Midjourney results may differ from the 2026 Gemini outputs in noticeable or subtle ways, depending on how each model's creators have adjusted training data, safety filters, or aesthetic objectives. The fact that the images are explicitly labeled as not representative underscores a growing awareness in AI research that visual generative models can perpetuate narrow or harmful stereotypes if consumed uncritically. The project's “just for fun” label does not absolve the underlying mechanism; it simply positions the exercise as an entertaining thought experiment rather than a definitive social study.
OpenAI said the ChatGPT Chrome extension upgrades are live in the beta channel. The broader rollout to all users is expected after the testing period, but no specific date has been announced beyond the beta window.
Valve said the new achievements features are available in the latest Steam client beta. The full rollout to all users will follow the beta period, with no announced release date beyond that window.