Google LLC today introduced a new image generation model, Nano Banana 2, that offers higher output quality and better performance than its predecessor. 

The company will make the algorithm available in more than a half-dozen products.

The original Nano Banana made its debut in August. Google followed up the model with an enhanced version, Nano Banana Pro, in November. The latter algorithm will continue to be offered after today’s launch. Google envisions customers using Nano Banana Pro for “high-fidelity tasks requiring maximum factual accuracy,” while the new Nano Banana 2 is better when the main priority is image generation speed. 

Google’s latest image model also brings several other improvements. According to the company, Nano Banana 2 is better than its earlier image generators at following prompts.

That means the model is less likely to leave out details requested by the user or make unnecessary additions. Furthermore, the model includes enhancements that will boost the quality of lighting effects and textures.

Many graphic design tasks require technical knowledge of the object being drawn. If a user requests an image that depicts a specific type of cloud, a neural network might have to draw on relevant metrological datasets. Nano Banana 2 uses information from the public web to understand users’ requests. 

“The model pulls from Gemini’s real-world knowledge base, and is powered by real-time information and images from web search to more accurately render specific subjects,” Google DeepMind product manager Naina Raisinghani wrote in a blog post. “This deep understanding also helps you create infographics, turn notes into diagrams and generate data visualizations.”

Consumers can also use Nano Banana 2 to generate image sequences. A designer, for example, could ask it to generate the first, last and middle frame of an animated scene. Google says that Nano Banana 2 is better than its predecessor at keeping design elements consistent across images. The model can “maintain character resemblance of up to five characters and the fidelity of up to 14 objects in a single workflow,” Raisinghani wrote in today’s blog post.

Users can customize the format of the images they generate. According to Google, Nano Banana 2 makes it possible to adjust the aspect ratio and specify a resolution from 512px to 4K.

Consumers can access the model via Google’s Gemini artificial intelligence app and Flow design tool. The model is also rolling out to several enterprise products including the Vertex AI suite of services in Google Cloud.

The launch comes about three months after OpenAI Group PBC released its newest image generation model. Like Nano Banana 2, GPT Image 1.5 is faster than its predecessor and follows prompts more accurately.

Image: Google