Google on Tuesday released Nano Banana 2.1, the newest version of its image generation and editing model, delivering wide-ranging improvements in visual quality, prompt adherence, text rendering and multi-turn character consistency while maintaining the Flash-level speed and cost efficiency of the product line. The model serves as the more efficient counterpart to Gemini 3 Pro Image, also known as Nano Banana Pro.

The update is rolling out across the Gemini app, Google Search's AI Mode, Google Ads and developer tools including Google AI Studio, Flow and Stitch. It arrives at a time when the Nano Banana family has become a central pillar of Google's consumer AI strategy. In September 2025, the original model's ability to turn selfies into collectible figurines propelled Gemini to the top of both major app stores, ending ChatGPT's nearly three-year dominance and coinciding with Alphabet's market value surpassing $3 trillion. Version 2 followed in February, built on Gemini 3.1 Flash Image, and introduced grounding in Google Search so that images of real events or people would be more accurate.

Google announced the release on X:

Meet Nano Banana 2.1, our latest image generation and editing model This upgraded version outperforms our previous models across the board, with notable leaps in visual design, mask-based editing, and subject consistency to help you create more natural-looking images. pic.twitter.com/d858DrzrF3

— Google (@Google) October 6, 2026

Sharper Editing and Stronger Benchmarks

Google highlighted three headline upgrades in the new release: improved visual design, more precise mask-based editing that changes only a marked region of an image, and stronger subject consistency so that a person or object remains recognizable across multiple edits. That consistency underpins multi-image workflows — storyboards, product sets, serialized artwork — where the alternative is retouching each frame by hand. On paper, the model handles up to 14 reference images simultaneously, maintaining consistency for up to four characters and fidelity for up to ten objects. It outputs at 1K, 2K and 4K resolutions, fixes tiling artifacts on extreme panoramic aspect ratios such as 1:4, 1:8, 4:1 and 8:1, and renders text and infographic layouts more accurately.

The gains are also visible in benchmarking. On overall preference in text-to-image tests, Banana 2.1 scored 1,050 ELO points, compared with 990 for Nano Banana 2 and 935 for Nano Banana Pro — a meaningful spread in a ranking system built from head-to-head human preference comparisons.

Developers gain additional control through configurable thinking levels, ranging from minimal to high, and through grounding with Google Web and Image Search, effectively allowing the model to look up information before generating an image.

The efficiency gains extend to cost. Through Google's developer API, a standard 1K image now costs $0.0336 — roughly half the $0.067 charged for Nano Banana 2 — while a 4K image runs $0.0756 versus $0.151 previously. Batch processing receives a further 50% discount. In practical terms, generating a thousand standard-resolution images now costs about $33.60, down from $67, positioning Nano Banana 2.1 as both a quality leader and one of the most cost-effective options in the generative image market. For products generating images in volume, per-image pricing is a direct input into operating costs, which is why the cut matters beyond the headline quality gains. How the benchmark gains hold up in day-to-day use as availability widens is the immediate test for Google's claims.

Source: Metaverse Post