For a long time, Google’s Imagen image generation engine had a big issue compared to competitors such as DALL-E—it could not properly generate images of people. Now, Google is finally opening up the option to all users again.
The free version of its Gemini chatbot is now gaining the ability to create images of people. This move follows the initial rollout of image generation to paying Gemini Advanced subscribers in an “early access” manner. Now, when you ask the chatbot to generate images that contain people, it will not evade your question anymore and will actually generate images with people in it.
I played around with it for a bit and found that it still has limitations. For one, it is only good for generating pictures of random people. When you ask it to generate a picture depicting someone identifiable, like a celebrity, it will attempt to generate it but it will ultimately shut down before it is done, telling you that it can’t generate images of “identifiable” people just yet. There’s an even harder limit on politicians and controversial figures where Gemini will not even try to generate an image in the first place, but that is also in place with most questions related to present-day politics. If your prompt is something like “make an image of a couple skydiving,” Gemini will do it—just don’t ask for any specific names.
For months, the ability to generate images depicting people was a key differentiator for Google’s premium Gemini Advanced service. It was launched all the way back in August 2024 as part of the Imagen 3 rollout, and among the many improvements Imagen 3 had, one of them was the model’s improved capability to handle queries that contain depictions of people. This feature was later extended to the Gemini side panel within popular Google Workspace applications like Gmail, Docs, and Slides, although you still needed to pay for the premium AI tier. Now, it’s slowly but steadily coming to free users as well—they still had image generation, just not with people, so it’s now less limited.
As a reminder, Gemini had previously shut down the ability to generate images of people because it had fundamental issues that Google needed to polish before it rolled out. Issues included overly sensitive diversity settings that would essentially randomize the race of people you were generating—which became a problem when it started creating images like a black George Washington. It could quickly become problematic, so Google opted to just add a hard filter to Gemini that would keep users from generating images with people in it until they sorted things out. It was initially supposed to be just for a few weeks, but it ended up taking months and the release of a brand-new generation model before we saw this again. Google probably underestimated how much work and re-training it would take to fix this.
The model still likely needs work, as evidenced by the fact that it still refuses some prompts, specifically those involving identifiable people. But hey, it’s better than nothing at all. And the images themselves look pretty crisp and realistic most of the time, too.
As of now, the new model is not available everywhere for free users, as Google is probably doing a scaled rollout of the feature to ensure absolutely nothing goes wrong with it. If you want to use it now and skip the line, you’ll still have to pay for Gemini Advanced, which would give you access to more stuff, such as experimental models, that you’ll find useful if you’re a frequent Gemini user.
Source: 9to5Google
