Recently, synthetic images created with 'Nano-Banana' have been uploaded daily on online communities and social media (SNS). Blurry photos are enhanced or backgrounds are swapped, and the results of naturally combining movie characters with real-life figures are shared within seconds. There is a reaction that 'Photoshop' (image editing tool) may not be needed, and the ability to solve tasks that used to take experts hours with just a single line of text commands is drawing attention from inside and outside the industry.
According to the industry on the 18th, Nano-Banana was introduced on the 14th via the Flux AI and FluxProWeb blog. It quietly appeared in the image editing platform LM Arena's 'Image Edit Arena' without any official launch or announcement from the developer, but it is gaining attention for its outstanding performance. Unlike before, there is no need to specify masks or layers; image editing is possible with just text commands like 'Please remove the car and add a sunset.' The result is modified to meet the requirements while maintaining the lighting, composition, and style of the original, leading to evaluations that it could 'replace Photoshop.'
The main functions of Nano-Banana include ▲sharpening blurry photos ▲combining two different images ▲replacing backgrounds ▲restoring obscured faces, among others. Its greatest strength is the ability to remember and reproduce the faces or characters from existing images. Even when requested from different angles or poses, the person's face remains largely unchanged and connects naturally, leading to assessments that 'AI synthesis artifacts' have significantly decreased. Thanks to this feature, there are also numerous cases on the internet where animation illustrations are converted into three-dimensional (3D) figure photos and shared.
Since Nano-Banana is not a formal version, users must press the 'Generate Image' button in LM Arena, upload the desired photo, and then enter prompts (commands). At this time, two AI models generate images simultaneously, and users select the better result in a battle-style operation. Since Nano-Banana is randomly assigned, it may not show up at once, and users may have to try several times to achieve the desired results. Users cite that this process not only makes the synthesis natural but also maintains the original face and atmosphere as the biggest advantage.
The true identity of Nano-Banana remains unclear. However, due to Google's precedent of using fruit names as codenames in internal projects, speculation is growing that it could be 'the next-generation Google image model.' Some are even predicting that it may be integrated into future Google AI Gemini or Pixel smartphones.
The rapid rise of Nano-Banana is set against the disappointment with OpenAI's GPT-5 model, which was released at the same time. The GPT-5 was first unveiled on the 7th (local time) and was promoted by Sam Altman, CEO of OpenAI, as 'doctor-level expertise,' but its actual performance has fallen short of expectations. There were errors, such as misnaming the first president of the United States and incorrectly writing state names on maps. Foreign media, including CNN, reported that 'GPT-5 has become a laughingstock,' and numerous posts mocking the erroneous outcomes flooded social media. As GPT-5 demonstrated underwhelming performance, some users even started a petition urging that they be allowed to 'use the previous version, GPT-4o,' which OpenAI eventually accepted.
An AI industry official said, 'In the fields of images and video, Google and other services continue to pursue new attempts, and results are accumulating rapidly,' and noted that 'as the focus of the AI market shifts from text to visual, the leading position of OpenAI could ultimately weaken.'