
In February, Google paused Gemini’s ability to generate images of people due to complaints about inaccuracies. You probably remember the black nazis etc..
For instance, when asked to depict “a Roman legion” the AI showed a diverse group of soldiers, and when tasked with “Zulu warriors”, it created stereotypical black figures.
Now, users on paid Gemini plans —Advanced, Business, or Enterprise— will be able to generate images of people once again, initially in an early access test, and only in English.
Google hasn’t said when it will be available for free users or in other languages.
Here’s what you should know:
Being able to generate people returns but only for paid users in early access.
Imagen 3, the latest model, aims to create fairer, more diverse images and will be available to all users.
Google’s latest model, Imagen 3, is designed to produce more “fair” images by improving the diversity in its training data. Google has been vague about the specifics but claims extensive testing has reduced the chance of undesirable results. Soon, all Gemini users will have access to Imagen 4, though people generation remains exclusive to premium tiers.
For this question I will close and resolve YES if we haven't been able to test this before the end of the year or haven't heard public debate on the matter before then (as I expect Google to delay public access if they are still having issues).
I will resolve to NO if everything seems as well as you expect (hard to replicate edge-cases are not counted, they will always exist in GenAI solutions) and we're able to get access to the tools before the end of the year ( I have paid Gemini access ).
As the question is hard to determine I will decide on the outcome (and not bet in this market) based on:
My job as head of an AI lab
Public discourse on medium, X, linkedin and several paid outlets like new york times etc.
Research papers published on arxiv in regards to "fairness" or "inclusivity" of images generated.
PS. As this is my first question, ever, I expect there to be interpretation issues, questions, noob errors by me, etc, but this is all done in best faith and I will not bet on this question myself. I have my opinions in the issue but I will do my best not to let them color my decision.