What did Google change with Nano Banana 2?

Faster, cheaper image generation aimed at wider use

Google upgraded its popular lightweight image model with a second‑generation release that focuses on speed, cost and broader availability. The new model—branded as the flash image variant of the company’s multimodal stack—prioritizes faster inference and lower production costs so it can be offered more widely, including to free users of the company’s AI assistants.

Key practical changes

• Performance and scale: The model runs faster than its predecessor, which reduces latency during image creation and makes it more practical for interactive applications on phones and web services.
• Output flexibility: It supports a wide range of resolutions, from social‑media‑friendly sizes to high‑resolution outputs, enabling use cases that span quick edits to larger downloads.
• Integration: The model is being folded into the firm’s broader assistant and image‑editing products as a default image generator, widening access across consumer and enterprise surfaces.

Why developers and businesses care

• Costs: By lowering compute needs per image, the update reduces the per‑request expense of deploying image generation at scale—an important barrier for companies wanting to add generative imagery to products.
• Productization: Making a faster image model the default inside client apps means more users will see and rely on generated images, which could accelerate adoption but also raises operational and moderation questions.
• Quality tradeoffs: The release emphasizes efficiency; enterprises that need the absolute best fidelity will still weigh higher‑capacity models against this faster, cheaper option.

Taken together, the release signals a pivot from research‑grade image models toward practical, production‑oriented tools that aim to make image generation routine across Google's apps and for a broader set of customers.