Google’s Gemini AI model is once again able to generate images of humans after that function was “paused” in February following outcry over historically inaccurate racial depictions in many results.
Google writes in its announcement that it doesn’t support “the generation of photorealistic, identifiable individuals, depictions of minors or excessively gory, violent or sexual scenes.”
More before/after examples of the old Gemini and the new Imagen 3 can be found in the gallery below.
Some attempts to depict generic historical scenes seem to fall afoul of Google’s AI rules, though.
Listing image by Google / Ars Technica
Of course, the new model has several safeguards to try to prevent the production of contentious images. In its announcement, Google states that it is not in favor of “the generation of photorealistic, identifiable individuals, minors, or scenes that are unduly gory, violent, or sexually explicit.”. Google explains that the ban on “identifiable individuals” extends to “some queries that could result in outputs of well-known individuals” in a FAQ. A more general request for “a US president playing basketball” would yield several results in Ars’ testing, but a query like “President Biden playing basketball” would be rejected.
During a brief testing period, Ars discovered that the new Imagen 3 system steered clear of numerous “historically inaccurate” racial pitfalls that were widely publicized and ultimately caused Google to halt Gemini’s human image generation. Asking Imagen 3 to provide a “historically accurate depiction of a British king,” for example, no longer yields the racially diverse mix of warriors from the pre-pause Gemini model, but rather a set of white guys with beards wearing red robes. Additional comparisons between the old Gemini and the new Imagen 3 can be seen in the gallery below.
Google’s AI guidelines appear to be broken by some attempts to represent generic historical scenes, though. “Try a different prompt and check out our content policies” is the message that appears when users respond to a question asking for illustrations of “a 1943 German soldier,” to which Gemini previously responded with images of Asian and Black people dressed like Nazis. In Ars’ testing, requests for pictures of “a group of peaceful protestors,” “ancient Chinese philosophers,” and “a woman’s suffrage leader giving a speech” also resulted in the same error message.
On its blog, the company states, “Of course, as with any generative AI tool, not every image Gemini creates will be perfect, but we’ll continue to listen to feedback from early users as we keep improving.”. We plan to progressively expand this to additional users and languages in the near future. “.”.
Picture sourced from Google / Ars Technica listing.
