Technical Specifications of Gemini 3.1 Flash Image Preview
| Item | Gemini 3.1 Flash Image Preview |
|---|---|
| Provider | |
| Model family | Gemini 3.1 (Flash tier) |
| Primary focus | Fast multimodal generation with image preview |
| Input types | Text, Image |
| Output types | Text, Image (preview generation) |
| Context window | Up to 1M tokens (Gemini 3.x Flash tier standard) |
| Latency tier | Low-latency, high-throughput |
| Streaming support | Yes |
| Tool calling | Yes (Gemini API tools framework) |
| Version | 3.1 |
What is Nano Banana 2
Nano Banana 2 is the popular nickname used by the press and developer community for the newly released Gemini-3.1-Flash-Image model. Google positions it as the “Flash”-tier image engine that brings near-Pro visual fidelity to a much lower latency and cost tier — suitable for high-volume generation, rapid iterative editing, and integrated product workflows across Google services. It inherits Gemini 3.1’s multimodal reasoning and adds image-centric capabilities (legible text in images, multi-image composition, wide aspect ratio support, native 4K).
Main features
- High-speed, multi-resolution generation: Flash-tier speed with options for 0.5K / 1K / 2K / 4K outputs and new extreme aspect ratios (1:4, 4:1, 1:8, 8:1).
- Real-time web grounding: Integrates both text and image search results to ground generated content in current web information when “Thinking” or search grounding is enabled. Useful for up-to-date references and factual infographics.
- Improved text rendering: Better short-text and graphic text rendering (fonts, sizes) than earlier Flash models; still imperfect on long paragraphs/small text.
- Multi-input editing and multi-turn workflows: Strong support for combining several images as inputs and for iterative edits across turns.
📊 Benchmark Performance — Image Generation & Editing (Elo scores)
| Capability | Gemini 3.1 Flash Image (Nano Banana 2) | Gemini 2.5 Flash Image (Nano Banana) | Gemini 3 Pro Image (Nano Banana Pro) | GPT-Image 1.5 | Seedream 5.0 Lite | Grok Imagine Image Pro |
|---|---|---|---|---|---|---|
| Text-to-Image — Overall Preference | 1079.0 ± 7.0 | 1073.0 ± 5.0 | 942.0 ± 6.0 | 1021.0 ± 5.0 | 1047.0 ± 5.0 | 928.0 ± 8.0 |
| Text-to-Image — Visual Quality | 1140.0 ± 6.0 | 1129.0 ± 6.0 | 929.0 ± 6.0 | 1043.0 ± 5.0 | 975.0 ± 5.0 | 759.0 ± 10.0 |
| Text-to-Image — Infographics (Factuality) | 1114.0 ± 14.0 | 1074.0 ± 12.0 | 881.0 ± 13.0 | 1102.0 ± 13.0 | 985.0 ± 12.0 | 890.0 ± 22.0 |
| Editing — General | 1065.0 ± 9.0 | 1047.0 ± 9.0 | 913.0 ± 9.0 | 1051.0 ± 10.0 | 995.0 ± 8.0 | 937.0 ± 9.0 |
| Editing — Character | 1056.0 ± 7.0 | 1049.0 ± 7.0 | 952.0 ± 7.0 | 1050.0 ± 8.0 | 1025.0 ± 7.0 | 894.0 ± 8.0 |
| Editing — Creative | 1023.0 ± 7.0 | 1031.0 ± 7.0 | 976.0 ± 7.0 | 1004.0 ± 7.0 | 1017.0 ± 7.0 | 938.0 ± 7.0 |
| Editing — Object/Environment | 1029.0 ± 8.0 | 1018.0 ± 8.0 | 945.0 ± 8.0 | 1042.0 ± 10.0 | 976.0 ± 8.0 | 946.0 ± 9.0 |
| Editing — Multi-Input | 1037.0 ± 8.0 | 1016.0 ± 8.0 | 919.0 ± 9.0 | 1056.0 ± 12.0 | 1014.0 ± 9.0 | N/A |
| Editing — Stylization | 1045.0 ± 7.0 | 1031.0 ± 7.0 | 862.0 ± 8.0 | 1045.0 ± 9.0 | 996.0 ± 7.0 | 984.0 ± 7.0 |
Key takeaways from this benchmark table:
- Across text-to-image generation and image editing categories, Gemini 3.1 Flash Image consistently leads or matches the highest scores among Flash-tier and many competitive image models.
- The model shows especially strong results in Visual Quality and Infographic (Factuality) benchmarks—signaling that it excels not only in aesthetic quality but also in rendering structurally accurate content.
- On Multi-Input editing, Nano Banana 2 also shows robust generalization, with higher scores than its previous Flash generation.
These evaluations are conducted via human side-by-side Elo comparisons on a diverse benchmark suite, reflecting both preference and fidelity across commonly used image generation/editing tasks.
Nano Banana 2 vs Nano Banana vs Nano Banana Pro
| Model | Positioning | Representative benchmark/notes |
|---|---|---|
| Gemini 3.1 Flash Image (Nano Banana 2) | Flash tier: speed + high visual quality (2K–4K) | Overall preference 1079.0 ± 7.0; visual quality 1140 ± 6.0 (internal GenAI-Bench). |
| Gemini 2.5 Flash Image (Nano Banana) | Earlier Flash release (lower fidelity) | Slightly lower preference/visual scores vs 3.1. |
| Gemini 3 Pro Image (Nano Banana Pro) | Pro tier: higher perceived fidelity for complex tasks, higher cost/latency | Different tradeoffs; some metrics show different relative rankings in specialty tasks. |
| GPT-Image 1.5 / other commercial models | Competitors (open/closed) | In Google’s internal benchmarks GPT-Image and others scored below Gemini 3.1 on visual quality and overall preference in the reported eval. Independent third-party comparisons vary. |
When to choose Flash Image Preview:
- Real-time image preview in apps
- Cost-sensitive large-scale image generation
- Interactive design assistants
How to access and integrate Nano Banana 2
Step 1: Sign Up for API Key
Log in to cometapi.com. If you are not our user yet, please register first. Sign into your CometAPI console. Get the access credential API key of the interface. Click “Add Token” at the API token in the personal center, get the token key: sk-xxxxx and submit.
Step 2: Send Requests to Nano Banana 2 API
Select the “gemini-3.1-flash-image-preview8” endpoint to send the API request and set the request body. The request method and request body are obtained from our website API doc. Our website also provides Apifox test for your convenience. Replace <YOUR_API_KEY> with your actual CometAPI key from your account. Where to call it:Gemini generates image
Nano Banana 2 supports image editing, image generation, and multi-image workflows. For image editing, you need to upload the image URL. For more parameters, please refer to the documentation.
Step 3: Retrieve and Verify Results
Process the API response to get the generated answer. After processing, the API responds with the task status and output data. You can directly download the image to your local machine in the playground (usually in PNG format). An image URL is generated in the API process; please download it promptly.