Nano Banana 2.1
An image generation model introduced by Google as part of the Gemini ecosystem. The model is available on the Gemini web and mobile clients and is positioned at a flash-tier level of image generation capability.
Core Facts
Timeline (last 90 days)
Nano Banana 2.1 can accept up to 14 reference images per task and can stably maintain the visual consistency of up to 4 characters and 10 independent objects across multiple rounds of rendering.
Nano Banana 2.1 is positioned as a cost-effective alternative for high-volume production, with competitive per-image prices for 2K and 4K output.
Nano Banana 2.1's native masked editing lets users hand-draw a mask or use a semantic prompt to change only the selected region.
According to feedback from multiple users, Nano Banana 2.1 is superior in the clarity and readability of text embedded in the image, with fewer spelling errors.
Nano Banana 2.1 fixes the tiling artifact problem under extreme aspect ratios.
Nano Banana 2.1 is positioned as an image model that strikes a balance between price and performance.
The official side gave three specific directions for Nano Banana 2.1's capability improvements: visual design, mask-based editing, and subject consistency.
Replicate currently already provides an access entry for Nano Banana 2.1, and developers can start testing it directly.
Nano Banana 2.1's official model card still notes limitations in character consistency.
Nano Banana 2.1 supports uploading up to 14 reference images.
17 more timeline events
All Facts (20)
Gemini 3.7 Flash calls Nano Banana when generating games to produce character and item textures as well as character animations.
65%UnverifiedNano Banana 2.1 is positioned as a cost-effective alternative for high-volume production, with competitive per-image prices for 2K and 4K output.
50%UnverifiedNano Banana 2.1 can accept up to 14 reference images per task and can stably maintain the visual consistency of up to 4 characters and 10 independent objects across multiple rounds of rendering.
50%UnverifiedNano Banana 2.1 fixes the tiling artifact problem under extreme aspect ratios.
50%UnverifiedAccording to feedback from multiple users, Nano Banana 2.1 is superior in the clarity and readability of text embedded in the image, with fewer spelling errors.
50%UnverifiedNano Banana 2.1's native masked editing lets users hand-draw a mask or use a semantic prompt to change only the selected region.
50%UnverifiedReplicate currently already provides an access entry for Nano Banana 2.1, and developers can start testing it directly.
50%UnverifiedNano Banana 2.1 is positioned as an image model that strikes a balance between price and performance.
50%UnverifiedThe official side gave three specific directions for Nano Banana 2.1's capability improvements: visual design, mask-based editing, and subject consistency.
50%UnverifiedNano Banana 2.1's official model card still notes limitations in character consistency.
50%UnverifiedNano Banana 2.1 supports uploading up to 14 reference images.
50%UnverifiedNano Banana 2.1 costs $0.0418 per image at the 1K/high setting, with an average generation time of 15.0 seconds.
50%UnverifiedThe final score across 8 head-to-head tests was OpenAI leading 3 to 2, with 3 draws.
50%UnverifiedNano Banana 2.1 has a watermark overlay bug: as the number of consecutive editing iterations increases, Google watermarks accumulate repeatedly, and image quality may degrade with each edit, exhibiting noise, color drift, and texture distortion.
50%UnverifiedAccording to Arena data, Nano Banana 2.1 improved by an average of approximately 61.7 points across three functions—text-to-image, multi-image editing, and single-image editing—compared to the previous generation Nano Banana 2.
50%UnverifiedGoogle has halved the price of 1K and 2K image generation for Nano Banana 2.1 and reduced the price of 4K images by approximately 25%.
50%UnverifiedNano Banana 2.1 ranks fifth in multi-image editing with a score of 1328, an improvement of 67 points; it ranks sixth in single-image editing with a score of 1428, an improvement of 38 points.
50%UnverifiedNano Banana 2.1 shows a marked improvement in Chinese text rendering compared to before, with improvements in Chinese text clarity and issues with incorrect characters.
50%UnverifiedImages generated by Nano Banana 2.1 can be handed off to Veo 3.1 within Google Flow to continue production and generate high-definition short clips, with the entire workflow completable within the same platform.
50%UnverifiedNano Banana 2.1 generates images quickly, but its image quality and overall level of polish still fall short of GPT (ChatGPT Images 2.5).
50%