Google Gemini Diverse Prompt Injection
Google Gemini Diverse Prompt Injection refers to the February 2024 discourse around Google's Gemini AI image generator producing images almost exclusively depicting people of color, even when prompts called for historically white figures like popes, 1800s Germans, or the founders of Fairchild Semiconductor. The behavior traced back to hidden diversity instructions injected into user prompts, sparking viral mockery on X and Reddit and forcing an apology from Google.
Overview
Google Gemini Diverse Prompt Injection refers to a February 2024 controversy in which Google's Gemini AI image generator produced images depicting people of color even when the user's prompt described historically white figures or settings, such as popes, 1800s Germans, or Scandinavian women1. Users quickly noticed that Gemini seemed to inject the word "diverse" and the phrase "people with diverse backgrounds" into the visible intro sentence of its outputs, revealing a system-level rewrite of the original prompt4.
The behavior drew comparisons to the earlier Ethnically Ambiguous AI Prompt Injection event and was widely described as "woke" by critics4. In one repeated test, Gemini refused to generate an image of a white person on the grounds that doing so would "reinforce harmful stereotypes," while it produced people of other races on request1. The gap between what users asked for and what Gemini returned became the entire joke of the meme.
How It Spread
How to Use This Meme
Posts in the format typically pair a straightforward prompt aimed at a historically white subject, such as a medieval European king, a Founding Father, or a 1940s German soldier, with Gemini's diverse output. The comedic beat is the mismatch itself, often shown as a four-image grid screenshot with the user's original prompt captioned above5.
Frequently Asked Questions
References (8)
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8