Nano Banana

Why It Exists

Users want to create AI-generated art and edit images from simple text prompts without uploading their data to the cloud or dealing with complex apps.

How It Works

Nano Banana leverages Google's Gemini Nano image generation models to create artwork from text descriptions entirely on-device on supported Pixel phones. The system can generate creative images, edit existing photos, apply artistic styles, and create personalized stickers from text prompts. All processing occurs locally using the Tensor NPU, ensuring privacy and eliminating the need for cloud processing. Nano Banana was introduced with the Pixel 10 series as an experimental AI art feature.

Everyday Use Cases

• AI artwork generation
• Photo style transfer
• Sticker creation
• Background replacement
• Creative concept visualization
• Personal avatar generation

User Workflow

• User opens Nano Banana in the Pixel Studio app
• Enters a text prompt describing the desired image
• Gemini Nano generates thumbnail options on-device
• User selects and refines the best result
• Final image is saved to gallery

AI Processing Flow

The text prompt is processed by on-device Gemini Nano language models to understand intent. A diffusion model generates the initial image on the Tensor NPU. Multiple candidates are created and ranked. The user selects a favorite for refinement. Additional on-device models enhance details and apply final adjustments. No data leaves the device.

Inputs / Outputs

Inputs:

• Text prompt
• Existing photo (for editing)
• Style preferences
• Color palette selections
• User refinement feedback

Outputs:

• Generated AI artwork
• Edited photos
• Custom stickers
• Style-transferred images
• Avatar designs
• Creative concepts

Known Limitations

• Only available on Pixel 10 series devices
• Limited to on-device model capabilities
• Generation time longer than cloud AI
• May not support complex multi-subject prompts
• Quality may vary for niche art styles

Unsupported Scenarios

• Professional graphic design
• Commercial use without attribution
• Non-Pixel devices
• Video generation
• Photorealistic human portrait generation

Performance Notes

Image generation completes in 15 to 30 seconds on Tensor G5; Thumbnail preview in 3 to 5 seconds; NPU acceleration for diffusion model; Battery consumption 10 to 15 percent per generation session

Available On

Shows where this feature is available and how its AI processing works on each platform. Availability may vary by device.

Research Status

Confidence and verification reflect how complete documentation is. Fields may show Not assessed or Not yet verified while research is ongoing - this flags gaps, not product deficiencies.

Research Notes

Available on Pixel 10a from launch in 2025; Requires Google Tensor G5 or later; Uses Gemini Nano 3 image model; Experimental feature with regular model updates; Supports 20+ artistic styles on-device

Details

Capability Mapping

This feature maps to canonical capability family: Image Generation.

Canonical capability: Image Generation.

• Image Generation - Creates novel imagery or emoji-like generated outputs from prompts.

Grounded in 1 device(s) on Gemini Nano.

Related Features

• Pixel Studio - Complementary