Nano Banana FAQ

Short answers about Nano Banana. Nano Banana is not open source. For a Nano Banana 2 local model or how to run the API client on your computer, see the local install guide. Getting-started steps are on the Quickstart. Comparisons are in Benchmarks.

Jump to: What is it How it works PC download Offline Model weights Speed Character consistency What to create Compared to others Commercial use API ChatGPT Image 2 Reproduce benchmarks
What is Nano Banana 2?
Nano Banana 2 is Google's name for Gemini 3.1 Flash Image. It is a hosted image generation and editing model that accepts text and image inputs. See Google's current model page for the supported capabilities and model ID.
Can I run Nano Banana locally?
No—the Nano Banana 2 model itself cannot be downloaded or run offline. You can run a small Python client on your computer, but that client sends the image and prompt to Google's hosted Gemini API. The step-by-step install is the Nano Banana 2 local model guide.

If you need a 100% offline solution (e.g., for air-gapped security), start with the local AI image generator guide and the Hardware Guide for models such as FLUX.1-dev, Qwen-Image, Z-Image, and HunyuanImage.
Can I download Nano Banana for PC/Windows?
No. There is no official Nano Banana desktop app, Windows installer, or .exe. Nano Banana is Google's hosted Gemini image model. Use it in the Gemini app or on the web, in Google AI Studio, or call the Gemini API from a script on your PC. The quickstart is the getting-started path, and the local guide covers that API client only. Unofficial “Nano Banana” .exe or APK downloads are not from Google and can be malware.
Does Nano Banana work offline?
No. Image generation runs on Google's servers and needs an internet connection. A Python script on your computer still sends the prompt and images to the Gemini API. You can download the images you generate, but the model itself does not run offline. For a separate model that can run without a network connection, start with the local AI image generator guide.
Can I download Nano Banana 2 model weights?
No. Google has not released Nano Banana 2 (Gemini 3.1 Flash Image) weights or training code. There is no checkpoint to download and run locally. The local guide installs the API client; inference stays in Google's cloud.
Is Nano Banana open source?
No. Nano Banana and Nano Banana 2 are Google's hosted Gemini image models. Google has not released the weights or the training code, so there is nothing to install or self-host. If you need an open-weight model you can run yourself, see Open Weights vs Open Source. The local install guide covers the API client, and the quickstart is the getting-started checklist.
Is it free?
The local Python client uses Google's Gemini API. As checked on August 18, 2026, Google's pricing page lists no free tier for gemini-3.1-flash-image. Pricing, quotas, and account availability can change. Open-weight models avoid that API fee, but hardware, electricity, storage, and cloud rental can still cost money.
How fast is it?
There is no honest universal number. Hosted API latency varies with the input, output size, network, service load, and account limits. Local models vary with checkpoint, precision, resolution, and hardware. Compare the exact workflow you plan to use instead of relying on a single advertised time.
How do I keep character consistency?
Google documents reference-image workflows and subject consistency, but no model guarantees an identical subject in every result. Keep the same high-quality references, describe which details must remain unchanged, and compare multiple outputs. See our prompt recipes for a practical checklist.
Can ChatGPT Image 2 run locally?
No public local weights are available. OpenAI documents the current product as GPT Image 2, an API model for image generation and editing. Calling that API from local code is not local inference.

If you want similar image generation quality that runs fully offline, we recommend checking out open-weight alternatives like Qwen-Image-Edit-2511 or HunyuanImage 3.0 in our Quickstart Guide.
Where is the API Key?
You can get your key from Google AI Studio. Once you have it, set GOOGLE_API_KEY or GEMINI_API_KEY in your terminal. The maintained request format is shown in Google's image-generation documentation.
Can I use images commercially?
It depends on the current service terms and the exact model or checkpoint. Check Google's applicable terms for Gemini API output, or the repository and model-card license for an open-weight model. Do not assume that one model family's license applies to every checkpoint, LoRA, demo, or hosted service.
How do I reproduce benchmark results?
Use the same disclosed input, prompt, model version, and settings where they are available. Hosted models can change, and both hosted and local inference may be nondeterministic across runs or hardware. We label missing inputs or parameters instead of calling an incomplete comparison reproducible.
Last fact-checked: