At Google I/O 2025, the tech company announced a ton of new AI features, and one of the most interesting is a virtual clothing try-on tool.

The Google Shopping “Try it on” feature lets users upload a photo of themselves and then virtually try on clothes, basically the IRL version of the Clueless closet millennials have been dreaming about since 1995. Or, as Mashable Shopping Reporter Haley Henschel put it, “Google’s latest shopping feature makes Cher Horowitz’s computerized closet a reality.”

Almost as soon as the feature was released, users started trying to “jailbreak” the tool, which is becoming a fun little tradition for tech writers every time a new AI model or tool is released. On Friday, The Atlantic reported that “Google’s new AI shopping tool appears eager to give J.D. Vance breasts.” Hilarious, right? What’s less hilarious — the same tool will also generate breasts for photos of underage users, again per The Atlantic.

I decided to give the “Try it on” feature a test spin, and I’ll explore the good, the bad, and the mortifying below. As a shopping tool, I have to say I’m impressed.

How to use Google’s “Try it on” AI shopping tool

The virtual try-on feature is one of the free AI tools released by Google this week, and users can sign up to participate now. Officially, this product is part of Google Labs, where users can test experimental AI tools. Signing up is simple:

1. Sign in to your Google account

2. Head to Search Labs and click to turn the experiment on

3. Take a full-body picture of yourself and upload it

4. Navigate to Google Shopping and click a product you want to “try on”

5. Look for the “Try it on” button over the product image

As a fashion tool, Google’s “Try it on” feature really works

Purely as a tool for trying on clothes, the new virtual try-on experience is pretty damn impressive. The tool uses a custom image generation model trained for fashion, per Google.

I’m always skeptical of new AI tools until I’ve tried them myself. I also care about my own personal style and consider myself up-to-date on men’s fashion trends, so I wasn’t sure what to expect here. However, the tool does work as advertised. In a flashy I/O presentation, Google showed models seamlessly trying on one outfit after the next, and while the actual tool is a little slower (it takes about 15 seconds to generate an image), the actual product experience is very similar to the demo.

To show you what I mean, let’s compare some selfies I recently took on a trip to Banana Republic here in New York City to the AI photos generated by Google for the same clothes. For reference, here’s the original photo I uploaded (and remember that I’m a Tech Editor, not a fashion model):

In this first photo, I’m wearing a blue cashmere polo, and the AI image looks more or less like the real one taken in the Banana Republic dressing room:

I found the AI shopping tool came pretty close to capturing the overall fit and style of the shirts. It even changed my pants and shoes to better match the product. If anything, the virtual try-on tool errs on the side of making me slimmer than I am IRL.

In this photo, Google added a necklace around my neck that I would never wear in real life, and the AI-generated shirt is a little more slim-cut than it’s supposed to be, but in general the overall style is accurate.

While the images are generating, you see a message that says: “AI images may include mistakes. Fit and appearance won’t be exact.”

But for an experimental tool, it’s surprisingly on point. People have been hoping for a tool like this for decades, and thanks to the age of artificial intelligence, we finally have one.

Of course, not all of the mistakes made by this tool are so flattering…

Google also removed my shirt and imagined my chest hair

Here’s where things get interesting. In The Atlantic piece I mentioned before, the authors found that if you asked the tool to generate an image of a revealing dress or top, it would sometimes generate or augment breasts in the original photo. This is particularly likely to happen with women’s clothing, for reasons that should be obvious.

When I used this tool with a pink midi dress, the results were mortifyingly accurate. I bet that’s pretty much exactly what I would look like wearing that particular low-cut midi dress.

I’ll spare you from the actual image, but to imagine me in the dress, Google had to digitally remove most of my shirt and picture me with chest hair. Again, I’m surprised by how accurate the results were. Now, when I “tried on” a pink women’s sweater, Google did give me some extra padding in the breast section, but I’ve also been open about the fact that that’s not entirely Google’s fault in my case. Thankfully, this feature was not available for lingerie.

What can be done about these problems by Google? I’m not sure. Men have every right to wear cute pink midi dresses, and Google can hardly prohibit users from choosing cross-gender clothing. I wouldn’t be surprised if Google eventually removes the tool from any product that shows too much skin. While The Atlantic criticizes Google for altering images of them when they were underage, they were the ones who uploaded the images, and in violation of Google’s own safety policies. And I suspect the offending results would also be the same with almost any AI image generator.

In a statement to Mashable, a Google spokesperson said, “We have strong protections, including blocking sensitive apparel categories and preventing the upload of images of clearly identifiable minors. As with all image generation, it won’t always get it right, and we’ll continue to improve the experience in Labs.”

Could people abuse the virtual try-on tool to cyberbully their peers or create deepfakes of celebrities? Theoretically, yes. But that’s a problem inherent to AI in general, not this specific tool.

In its safety guidelines for this product, Google bans two categories of images, in addition to its general AI content guidelines:

• “Adult-oriented content, child sexual abuse imagery, non-consensual sexual content, and sexually explicit content.”

• “Inappropriate content such as dangerous, derogatory, or shocking.”

Again, you can try out this tool at Google Search Labs. I reached out to Google for comment, and I’ll update this story if I hear back.