Google’s latest visual AI model, named Nano Banana Pro, generates highly realistic images that often resemble photographs taken with a smartphone. Users have noted that, despite some imperfections—like oddly rendered backgrounds—many images could easily be mistaken for real photos when viewed in passing. This increasing realism raises questions about how AI-generated images can be distinguished from genuine photographs.
The model’s ability to achieve this likeness can be attributed to several factors, such as exposure and depth of field typically associated with smartphone photography. Visual cues that usually indicate digital generation, like unusual textures or lighting, are becoming less apparent in the outputs of Nano Banana Pro. Experts, including Ben Sandofsky, co-founder of the iPhone camera app Halide, suggest that the model incorporates elements like aggressive image sharpening, commonly seen in smartphone images.
Interestingly, Nano Banana Pro does not use Google Photos as its source for imagery. Instead, it connects to Google Search, allowing it to include real-world information contextually, which helps it generate elements that naturally fit within the scenes it creates. This includes features like historical clothing and logos that enhance the authenticity of the images.
Despite improvements, the potential for AI-generated content to produce “hallucinations” remains. For instance, the model might create misleading or incorrect outputs, like outdated watermarks or other unintended details. As the technology advances, users find it increasingly challenging to discern AI-created images from real ones, prompting concerns about the implications for trust in visual media. This situation suggests that vigilance regarding digital imagery may be more necessary than ever.
Source: https://www.theverge.com/report/837971/google-nano-banana-pro-realistic-phone-photos