Gemini's AI saree trend goes viral on Instagram – but are your photos safe?

Recently, social media has been buzzing with a new trend involving AI-generated vintage saree portraits, fueled by Google's innovative Nano Banana feature. This AI photo-editing tool transforms selfies into whimsical 3D figures, characterized by glossy, toy-like aesthetics with exaggerated facial features. As users flock to create these unique saree images, typically showcasing women in traditional attire against nostalgic backdrops, the phenomenon has drawn both fascination and caution regarding digital safety. Despite claims from tech giants like Google and OpenAI regarding the safety measures for shared content, experts warn that individual practices and the intent of those who access these images play a crucial role in protecting privacy. Google's Nano Banana feature incorporates an invisible digital watermark, known as SynthID, which, along with metadata tags, helps identify AI-generated content. According to information available on aistudio.google.com, all images edited or created with the Gemini 2.5 Flash Image include this watermark, aiming to promote transparency and trust among users. While the SynthID watermark is designed to ensure authenticity, its invisibility means that most users cannot verify its presence without specific detection tools, which are not publicly accessible. This limitation raises questions about the reliability of watermarks in preventing misuse of images. Critics point out that these digital markers can be manipulated, removed, or overlooked, as noted by Ben Colman, the CEO of Reality Defender, who remarked on the shortcomings of current watermarking applications. Experts suggest that watermarking should not be the sole method of protection. UC Berkeley professor Hany Farid emphasized the need for a multi-faceted approach to combat the creation of convincing fake images. To enhance personal safety when engaging with viral trends, users are advised to exercise caution by avoiding the upload of sensitive images, stripping metadata before sharing, adjusting privacy settings, retaining original files for comparison, and thoroughly reviewing platform terms regarding image rights and AI training usage.