Gemini AI saree photos look fun on Instagram, but are they putting you at risk?

AI tools like Google Nano Banana have safety features, however, user intent and our own smart safety practices are key to preventing misuse.

If you’ve been online on Instagram recently, chances are you’ve either tried out the Google Nano Banana trend and the viral vintage saree AI edits, or at least stumbled upon them while scrolling through your feed or reading about them elsewhere.

The vintage saree AI trend reimagines portraits in retro-inspired looks.

The “Nano Banana” craze comes from an AI photo-editing feature built on Google’s Gemini Nano model. It turns ordinary selfies into striking 3D figurine-style portraits, complete with glossy plastic-like skin, oversized expressive eyes, and playful cartoonish proportions. While experimenting with this tool, users also sparked another wave of creativity—the vintage saree AI trend, which reimagines portraits in retro-inspired looks.

The vintage saree AI-generated portraits feature photos of users – mostly women – turned into elegant images showing them in traditional sarees, often in stylised cinematic or vintage backdrops.

As usual, with any AI trend comes concern about safety and privacy of our photos.

How safe is Gemini Nano Banana tool?

Invisible watermark: SynthID

While tech companies like Google and OpenAI (maker of ChatGPT) offer tools to safeguard user-uploaded content, it's ultimately our own safety practices – and the intent of those accessing the images – that determine how likely the content is to be misused, modified without consent, or falsely attributed.

Google's Nano Banana images carry an invisible digital watermark (SynthID), plus metadata tags, which are meant to identify content as AI-generated.

“All images created or edited with Gemini 2.5 Flash Image include an invisible SynthID digital watermark to clearly identify them as AI-generated. Build with confidence and provide transparency for your users,” information on aistudio.google.com reads.

While there always is a concern over AI-generated images looking too real, making photos prone to deepfakes, the SynthID digital watermark – even though not visible to naked eye – can confirm that AI played a role in creating or editing the content when analysed with specific detection tools, according to spielcreative.com.

This gives individuals and platforms a method of verifying an image’s origin.

However, the detection tool for that watermark is not yet publicly available. So while the watermark exists, most everyday viewers cannot verify it, as per a Tatler Asia report.

Watermarking at first sounds like a noble and promising solutions but its real-world applications fail from the onset when they can be easily faked, removed or ignored, a report by Wired quotes Ben Colman, CEO of AI-detection startup Reality Defender.

Some experts think watermarking can help in AI detection but its limitations need to be understood. Nobody thinks watermarking alone will be sufficient," Wired quotes Hany Farid, a professor at the UC Berkeley School of Information who thinks improving upon the tool and using it with a combination of other technologies can help make it harder to make convincing fakes.

How to keep your photos safe

Be selective about what you upload: As with any AI tool, what you get out is only as safe as what you put in. Avoid sensitive photos (private, intimate, or high-value personal identifiers) as much as possible.

Strip metadata: Before uploading, details like remove location tags, device information etc. This can help in preventing leakage of unintended information.

Privacy settings: It is always ideal to use strong privacy settings on apps and social media platforms. Controlling who can see your images can prevent content from being misused or getting accessed by fraudsters. Be cautious before sharing widely as once shared publicly, images can be copied, altered, used out of context.

Retain copies: Keep your original image or prompt backup so you notice changes or misuse.

Watch for terms & consent: Understand whether uploading gives the platform rights over your image, whether it can be used in model training etc.