Frequently Asked Questions

Clear, practical answers about Nano Banana. Need more help? See Quickstart or explore Benchmarks.

Jump to: What is it How it works 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.

If you need a 100% offline solution (e.g., for air-gapped security), check out our Hardware Guide for alternative models like Z-Image, Qwen-Image, and HunyuanImage.
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: