Gemini Image Generation (Nano Banana)
The Gemini Image Generation (Nano Banana) workspace (/nano-banana) provides a rich studio for text-to-image synthesis, multi-image reference conditioning, conversational image editing, and real-time Google Search Grounding using the Gemini native image generation family.
Source Architecture
Section titled “Source Architecture”| Layer | File Path | Responsibility |
|---|---|---|
| Mesop Page & State | pages/gemini_image_generation.py (GeminiImageGenState) |
Studio UI, multi-image reference uploader (up to 14 images), aspect ratio & output size controls, Search Grounding toggle, and Quick Prompt presets. |
| Model Adapter | models/gemini.py (generate_image_from_prompt_and_images) |
Builds multimodal types.Part contents, configures ImageConfig (aspect_ratio, image_size) and ThinkingConfig, invokes client.models.generate_content(), and stores generated PNGs to GCS. |
| Model Registry | config/gemini_image_models.py (GEMINI_IMAGE_MODELS) |
Declarative capability specification (aspect ratios, output sizes 1K/2K/4K, max reference images, and Search Grounding support) per Nano Banana variant. |
| Prompt Presets | config/banana_presets.json |
Curated library of creative prompt templates and multi-image composition patterns. |
Supported Models (config/gemini_image_models.py)
Section titled “Supported Models (config/gemini_image_models.py)”| Display Name | Model ID | Max Reference Images | Output Sizes (image_size) |
Aspect Ratios | Search Grounding |
|---|---|---|---|---|---|
| Gemini 3.1 Flash Image Preview (Nano Banana 2) | gemini-3.1-flash-image-preview |
14 | 1K, 2K, 4K |
1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9, 1:4, 4:1, 1:8, 8:1 |
Yes |
| Gemini 3.1 Flash Image Lite Preview (Nano Banana 2 Lite) | gemini-3.1-flash-image-lite-preview |
14 | 1K, 2K, 4K |
All 14 ratios (1:1 to 8:1) |
Yes |
| Gemini 3 Pro Image Preview (Nano Banana Pro) | gemini-3-pro-image-preview |
14 | 1K, 2K, 4K |
Standard 10 ratios (1:1 to 21:9) |
Yes |
| Gemini 2.5 Flash Image (Nano Banana) | gemini-2.5-flash-image |
3 | 1K, 2K |
Standard 10 ratios (1:1 to 21:9) |
No |
Key Capabilities
Section titled “Key Capabilities”1. Multi-Image Reference Conditioning & Conversational Editing
Section titled “1. Multi-Image Reference Conditioning & Conversational Editing”- Users can upload up to 14 reference images (with Nano Banana 2 / Pro) or select existing images from the Studio Library.
- Any generated image in the output gallery includes a “Continue Editing” action that promotes the generated image into the reference slot so users can iteratively refine lighting, outfits, backgrounds, or typography across turns.
2. Ultra-Wide & Panoramic Aspect Ratios + 4K Sizing
Section titled “2. Ultra-Wide & Panoramic Aspect Ratios + 4K Sizing”- Nano Banana 2 (
gemini-3.1-flash-image-preview) expands beyond standard photographic ratios to support extreme banner and vertical ratios (1:4,4:1,1:8,8:1) alongside native1K,2K, and4Koutput resolution settings (types.ImageConfig(aspect_ratio=..., image_size=...)).
3. Google Search Grounding
Section titled “3. Google Search Grounding”- When Enable Google Search Grounding is toggled on (
use_search=True),models/gemini.pyattachestools=[types.Tool(google_search=types.GoogleSearch())]to theGenerateContentConfig. - This allows Nano Banana to ground visual details in real-world entities, current events, landmarks, or accurate product designs before synthesizing the image.
4. Thought Capture & C2PA Provenance
Section titled “4. Thought Capture & C2PA Provenance”- Model reasoning thoughts (
part.thought == True) are separated from final text descriptions and persisted on the FirestoreMediaItem(thoughts), alongside any embedded C2PA Content Credentials extracted from the generated PNG bytes.