I tested Nano Banana 2.1 against ChatGPT, and it was close

Google recently upgraded its Nano Banana AI image generator, now available inside Google Gemini. In a test against ChatGPT's image generator, Images 2.5, the results were close but not conclusive. Both systems produced high-quality images capable of realistic lighting, glass reflections, and intricate details.

One prompt asked the models to create a photorealistic scene of a Victorian greenhouse inside a glass snow globe during a thunderstorm. ChatGPT featured an elaborate, golden-lit greenhouse, while Nano Banana 2.1 opted for a more restrained design with metal framing and warm interior lighting. The Nano Banana image's reflections and lighting appeared more natural, though the wet tabletop was unexpected.

Another test involved generating a retro-futuristic travel poster for Saturn's moon Titan, requiring accurate text rendering. ChatGPT's poster looked luxurious, but it left a gap between the $ and the 2 in the price. Nano Banana 2.1 produced a more cartoonish but accurate poster, including the requested text and price.

For character consistency, the models were tasked with creating a four-panel comic showing an orange fox discovering a magical key, unlocking a door, and entering a library. ChatGPT's illustrations had a children's book style, while Nano Banana 2.1 produced a more traditional comic strip with bold outlines and consistent character appearance. Google's version felt like a cohesive illustrated story, whereas ChatGPT's panels seemed more independent.