Google says its new AI can generate images that look more like you. I tried it, and I’m both impressed and terrified at the same time

Cue the return of the AI action figure trend: Google has just released a new AI photo model with the ability to remix photos and generate images from an inspiration photo. The update gives users the ability to reimagine themselves as nearly anything they can type into a prompt box. A key feature? Google says Gemini can now generate images that actually look like you.

Google says that the new Gemini Flash 2.5 is designed to allow users to create photos that more closely resemble images of themselves or their friends. The update also allows users to remix photos together, asking the AI to combine two images and a text prompt, like reimagining a butterfly’s wings as the pattern of a dress.

I wanted to see what the new Gemini model was capable of, so I quickly opened Gemini, typed in a few prompts, and found myself staring at generated images of myself that had me wavering between being thoroughly impressed – and incredibly creeped out.

The first thing that came to mind when I read about Gemini’s “Reimagine this person as” feature, I immediately thought of the viral AI action figure trend, so I uploaded the same photo that I use for my writer’s bio and asked Gemini to turn me into an action figure, naturally with a laptop and a camera as accessories.

I have to admit, I was not expecting to see something that looked so close to my likeness wrapped in plastic. Impressively, the AI generation looks quite a bit like me and even copies the same outfit from the inspiration photo. The resemblance is impressive, but it also felt just a bit creepy. The text is legible, albeit mispelled.

The AI strangely gave my action figure a second pair of hands, which served as the inspiration for the second prompt that I typed in. As a mother and full-time editor with a photography business side gig, I never feel like I have enough hands (or time). I asked the AI to depict me with a few extra hands trying to manage it all.

The result looks obviously AI-generated, yet I still do rather look like a cartoonish version of myself.

I then tried to reimagine myself in careers I may have had in another life, as a wildlife photographer and in another location photographing the Northern Lights in Iceland. I don’t really think these generations look much like me at all – she has glasses and a similar skin tone, but nothing else feels quite right. Even the hair is too long. (And why is my camera pointed into the trees when there's a bird six inches from my face?)

Image 1 of 2

With a mix of curiosity and trepidation, I asked Gemini to reimagine what I may look like when I’m 70. I have to admit, with the nose and eyes, I think it may be a fairly accurate guess, and the woman in the generation even reminds me a bit of my grandmother.

In order to recover from that experience, I had to ask the AI to imagine what I would look like as a Gen Z teenager. I look more like a younger sister than myself, but it’s not terrible. (Perhaps I should pick up some ripped jeans and combat boots.)

I did try some more prompts that could have more practical use rather than for entertainment. I told the AI I wanted to change my hair and asked for a few different options. (For the record, I do not look good as a blonde.)

While many of the AI generations had me both impressed and creeped out, one in particular took the cake in making an impression, not because the results were good, but because it suggests the AI isn’t completely biased.

The photo that I uploaded is just a head and shoulders shot with a closed smile; the AI can’t really tell my body shape or the condition of my teeth from the reference photo. One of the generations showed a person who was not a size zero and who had a bit of a charming tooth gap. The fake Hillary’s body shape and teeth don’t match mine, but it hints that the reimagine feature isn’t going to try to fill in missing information with a Photoshop-perfect stereotype.

Some would probably be offended, but I found it to be a bit the opposite. AI is trained on thousands of different images, and the images that get fed into the training data can create biases, leading the AI to think that people should look a certain way. One generation isn’t really enough to rule out potential biases, but it does hint that Gemini’s training data isn’t made up of one body shape. Women have enough body issues without even fake AI generations contributing to the stereotype.

Training data is where much of the debate around the use of AI comes in. Gemini assumes that you have the rights to the photos that you upload. Gemini told me, “It is your responsibility to ensure you have the necessary rights to any content you upload.” There appears to be no built-in protection against downloading an image from the internet and uploading it to the AI for remixing.

Google also doesn’t share where it sources the images for training, and the company has been embroiled in legal battles over using intellectual property to train its AI.

Gemini does, however, add a watermark to the metadata to indicate that the resulting images are AI creations.

You may also like

Before you go, learn how to spot an AI-generated image.