A significant shift is happening within Gemini’s image generation capabilities. Nano Banana 2 is rapidly replacing Nano Banana Pro across the platform, including within AI Mode for search and Google Lens. Every image created will continue to be subtly marked with Google’s SynthID, a digital signature confirming its AI origin.
The change unlocks a substantial benefit for free users. Previously, access to Nano Banana Pro was severely limited, sometimes allowing only a handful of image generations per day. Nano Banana 2’s increased efficiency dramatically expands creation possibilities, delivering quality nearly on par with the Pro model – though usage limits may shift over time.
For those with paid subscriptions, Nano Banana Pro remains available, but Nano Banana 2 will be the default option. This indicates Pro still holds a slight advantage, particularly for tasks demanding “high-fidelity” and “maximum factual accuracy,” where intricate detail is paramount.
Initial testing reveals Nano Banana 2 isn’t a revolutionary leap, but a noticeable improvement in speed. The difference isn’t always immediately apparent, but users will experience quicker image generation. Extensive testing with diverse prompts has proven its impressive capabilities, even if perfection remains elusive.
The Gemini app now features a new template system. Users can select from presets like “Gothic clay” or “Oil painting” to refine their prompts, or simply create images from scratch. Existing functionality to upload starting images, using the plus (+) button, remains available.
Subscribers to Gemini’s AI Plus, AI Pro, or AI Ultra plans have a unique option: after Nano Banana 2 renders an image, they can select “Redo with Pro” to generate the same prompt using Pro’s enhanced processing power. Note that selecting this option appears to reset any chosen template.
Testing involved complex requests, such as generating a papercraft-style weather forecast for New York, combining real-time data, image creation, and text rendering. Both models delivered high-quality results, though the Pro version exhibited slightly more detail, as expected.
Further tests explored landscape painting and infographic creation, even venturing into a comic strip adaptation of Charles Dickens’ *Bleak House*. The results were largely impressive, showcasing accurate text, stylistic adherence, and minimal errors.
Despite the advancements, AI-generated images aren’t flawless. Occasional misspellings, physics inconsistencies, and a generic aesthetic – stemming from training on vast datasets – still appear. These models are still learning to truly create something original.
Direct comparison between Nano Banana 2 and Nano Banana Pro is challenging, as “Redo with Pro” generates a completely new image. However, Pro generally maintains an edge in quality and nuanced details, though Nano Banana 2 occasionally produced preferable results.
The most significant current benefit of Nano Banana 2 lies in Gemini’s ability to integrate web-sourced information – like weather conditions – into imagery. The models are also becoming increasingly adept at mimicking diverse styles, and leaving behind fewer telltale signs of AI creation.
