Google’s Nano Banana 2.1 adds 4K editing, at half the price, with caveats

Google rolled out Nano Banana 2.1, the latest version of its image generation and editing model, across the Gemini app, Flow, and Google AI Studio. The update leans hard into photorealism rather than raw novelty: outputs now go up to native 4K in aspect ratios as wide as 8:1, and Google says the lighting and texture quality is noticeably less synthetic than earlier versions, closer to what the company describes as television-level sharpness than the slightly plastic look that’s dogged AI image models for years.

The bigger functional changes are in editing rather than generation. Mask-based edits are meant to hold up across repeated passes instead of degrading the source image the way earlier models would after a few rounds of touch-ups, and the model can track up to 4 people and 10 objects across a scene while handling as many as 14 reference images at once, which matters for anyone trying to keep a consistent cast of characters or product shots across a batch of images. According to Google’s own model card, though, this isn’t a fully solved problem yet: the company lists partial instruction-following on masked and doodle-based edits as a known limitation, and says character consistency between a reference image and the generated output “is not always perfect.” The improvement is real on benchmarks, just not the clean fix the marketing framing implies.

Pricing dropped by roughly half alongside the quality bump. A 1K image now costs about 3.36 cents, down from 6.70 cents on Nano Banana 2, and a 4K image runs 7.56 cents instead of 15.10. Cutting the price while raising the resolution ceiling is an unusual combination for a model update, and it puts pressure on competing image editors that haven’t matched both moves at once.