Google's Nano Banana 2.1 Outperforms Pro on Image Edits

Google has introduced Nano Banana 2.1, an upgraded version of its speed- and cost-optimized image generation and editing model. Available via the API as gemini-nano-banana-2.1, the new model is built on Gemini 3.6 Flash. It introduces a configurable Thinking setting with minimal, medium, and high options to balance latency and quality. The rollout spans Google AI Studio, the Gemini app, Google Flow, Stitch, Google Ads, and Search AI Mode.

According to Google's internal human-evaluation Elo benchmarks, Nano Banana 2.1 outperforms both Nano Banana 2 and the higher-tier Nano Banana Pro on most image-editing tasks. The model also achieves an infographic factuality score that is roughly 2.9 times higher than its predecessor, largely driven by its ability to ground generations using Google Web and Image Search results.

The model supports 1K, 2K, and 4K output resolutions, with the higher resolutions designed to reduce tiling artifacts and improve realism. For complex compositions, it can ingest up to 14 reference images, allowing developers to maintain consistency for up to four characters and preserve the details of up to 10 objects. These features target workflows in visual composition, mask-based editing, subject consistency, and rendering quality.

Despite these gains, Google noted several production limitations. The model can struggle with spatial reasoning, such as confusing left and right, and small text may appear blurry at 1K resolution. It also faces occasional pose leakage during edits and character appearance drift. However, because pricing and latency remain aligned with the budget-friendly Nano Banana 2, practitioners can now use gemini-nano-banana-2.1 as a faster, cheaper primary candidate for editing and character-driven campaigns before routing to Pro.