Nano Banana 2.1 Image Generator (Google)

Overview

Nano Banana 2.1 is Google’s image generation and image editing model in the Gemini family. You describe what to create or change, and it produces a finished image in 1K, 2K, or 4K sizes.

It can also work from reference images. You can fuse multiple references into one result and keep key subjects consistent across iterations. This makes it useful for designs that need readable text, repeatable characters, and controlled layouts.

What you can build

• Product mockups that keep a brand object consistent across scenes
• Character sheets and multi-panel story images with the same person shown repeatedly
• Posters, magazine covers, and thumbnails with readable titles and labels
• Infographics that mix diagrams, icons, and short text blocks
• Style transfers where you apply the look of a reference photo to a new scene
• Photo edits like background swaps, lighting changes, and object replacements
• Composites that combine multiple photos into one coherent image

Inputs

• A required written instruction that describes the scene you want, or the exact edit you want applied
• Optional reference images to guide the result or to edit directly. You can provide up to 14 images in one request, including up to 10 object references and up to 4 character references for consistency.
• A target output size selection (1K, 2K, or 4K)
• An optional aspect ratio selection. If you don’t set one, the model will pick an appropriate frame, often matching an input image when provided.
• An optional random seed number if you need more repeatable outputs between runs
• A chosen output file type for the generated image (JPEG, PNG, or WebP)
• An optional safety filter strictness level that controls how aggressively the system blocks sensitive content

Outputs

• One generated image file in the format you selected (JPEG, PNG, or WebP)
• Pixel dimensions that follow a fixed set of size-and-ratio combinations

Supported aspect ratios and pixel sizes:

Aspect ratio	1K output	2K output	4K output
1:1	1024×1024	2048×2048	4096×4096
1:4	512×2048	1024×4096	2048×8192
1:8	384×3072	768×6144	1536×12288
2:3	848×1264	1696×2528	3392×5056
3:2	1264×848	2528×1696	5056×3392
3:4	896×1200	1792×2400	3584×4800
4:1	2048×512	4096×1024	8192×2048
4:3	1200×896	2400×1792	4800×3584
4:5	928×1152	1856×2304	3712×4608
5:4	1152×928	2304×1856	4608×3712
8:1	3072×384	6144×768	12288×1536
9:16	768×1376	1536×2752	3072×5504
16:9	1376×768	2752×1536	5504×3072
21:9	1584×672	3168×1344	6336×2688

Limitations

• The model only outputs the supported aspect ratios listed above. It won’t produce arbitrary pixel sizes.
• Using many reference images can reduce control. Details may drift when references conflict.
• Extreme aspect ratios (like 8:1 or 1:8) can be harder to compose. Small subjects may look cramped.
• Low-quality inputs can degrade results. Blurry photos, heavy JPEG compression, tiny on-image text, and scanned screenshots often cause artifacts.
• Safety filtering can block prompts, reference images, or outputs. When that happens, you may get no image back.
• A seed can improve repeatability, but it can’t guarantee identical images across model updates.

Safety & compliance

• Follow Google’s content rules and your local laws when generating or editing images.
• Don’t request sexual content involving minors, explicit sexual content, or instructions for wrongdoing.
• Don’t use the model to create deceptive impersonations. Get consent before using a real person’s likeness.
• Respect copyrights and trademarks. You’re responsible for how you use generated assets.
• If you need stricter moderation, raise the safety setting. This can reduce risky outputs but may block more requests.