Google’s Gemini Nano Banana: AI tool that creates realistic photo edits
Google’s Gemini app now features Nano Banana, an AI image generator powered by Imagen 4. It creates lifelike edits, preserves subject identity, and even generates realistic text in photos.
Google has recently introduced its latest AI-powered image generation and editing tool, informally known as Nano Banana, inside the Gemini app. Within days of launch, it has gone viral, with users creating 3D figurines, retro saree portraits, 'Hug Your Younger Self' portrait and celebrity polaroids. While AI image editing is not very new, Nano Banana stands out with its ability to deliver lifelike details and preserve the original likeness of people and objects.
What is Nano Banana?
Nano Banana is Google DeepMind’s new image generator and photo editor inside the Gemini app. It can:
- Generate new images from text prompts.
- Edit existing photos by adding or removing objects.
- Combine multiple pictures into one.
- Maintain consistency of faces, pets, and objects across edits.
The technology behind it
At its core, Nano Banana is powered by Imagen 4, Google DeepMind’s latest text-to-image model launched in May 2025. Imagen began as a Google Brain project and was later developed under DeepMind after the 2023 merger. This collaboration accelerated research, leading to sharper image quality, realistic lighting, and improved text rendering.
How does the Imagen 4 work?
Imagen 4 uses two main techniques:
- Large Language Models (LLMs) like Google T5 are used to understand text prompts.
- Diffusion models that refine random noise step by step into a polished image.
The process starts with a low-resolution image and gradually “upsamples” it to 2048x2048 pixels, making images sharper and more detailed.
Why Nano Banana looks more realistic than ChatGPT’s Ghibhi
Unlike older AI tools, Nano Banana emphasises editing accuracy without losing key details. It keeps perspective, lighting, and subject identity intact. For example, when blending two photos, the final image still preserves recognisable features of both.
It is also better at generating text within images, such as signboards, labels, and packaging, an area where most AI models often fail.
Features for users
With Nano Banana, users can:
- Generate images from scratch.
- Edit photos seamlessly.
- Blend multiple shots into one.
- Receive up to four variations of an image in one request.
Ensuring authenticity with SynthID
To tackle concerns about misuse, Google adds an invisible SynthID watermark in every AI-generated image. This watermark survives common edits like cropping or resizing.
Furthermore, Google uses Content Credentials (C2PA) to attach tamper-proof metadata about when and how an image was created. A SynthID Detector tool is also being tested for public use.
Nano Banana showcases how Google is trying to lead the AI image race with realistic edits and strong authenticity measures. However, as generative AI spreads, it also raises concerns about misuse, from fake images to misleading edits. The tool is now available via the Gemini app, Gemini API, and Vertex AI for enterprises.