You do not need a paid cloud project to touch Nano Banana 2. AI Studio’s free tier is enough for UI tests and a first API image—know the limits before you wire production.

UI path (two minutes)

Open Google AI Studio and sign in.
Model picker → Gemini 3.1 Flash Image (Nano Banana 2 family).
Prompt: Product photo of a stainless water bottle on white, soft shadow, 4:5.
Download or iterate with follow-up edits.

Watch rate limits: free tiers throttle bursts; batch jobs belong on Vertex with billing.

API path (developers)

AI Studio → Get API key (do not commit to Git).
Use the image-capable Gemini model id from current docs (e.g. gemini-3.1-flash-image-preview—verify in Studio before shipping).
Send multimodal generateContent with your text prompt.

Pseudo-flow:

# Store key in env — never in repo
export GEMINI_API_KEY="your-key"

Your app should read the key from environment variables, set aspect ratio in the prompt or config, and save binary image bytes to disk.

Explore Nano Banana Workflows

Prompt tips that transfer from UI to API

One scene per request for reliability.
Quote text literals.
Ask for aspect ratio in natural language if the SDK lacks a knob.

Security

Keys in .env.local only; rotate if leaked. MidassAI site builds must not embed keys—this tutorial is for your local tooling.

Next step

Once the first call works, wrap prompts in your CMS or MidassAI Nano workflows so marketers never touch curl.

Prerequisites and setup

You need a Google account with AI Studio access enabled to proceed. Standard consumer accounts work for testing, but organizational policies might block external API key generation. Ensure you have a code editor like VS Code installed to manage environment variables securely on your local machine. A terminal capable of running curl commands helps verify API responses before integrating them into larger applications.

Assume the model identifier remains gemini-3.1-flash-image-preview during the beta period. Google frequently updates endpoint names, so check the model picker weekly for changes. If you plan to automate generation, install the latest Python SDK or have a HTTP client ready for direct calls. Do not rely on deprecated v1 endpoints for image generation tasks.

Extended prompt workflow

Define lighting conditions explicitly in your text string. Instead of vague terms like "soft shadow," specify key light at 45 degrees, fill ratio 1:2 to reduce variance between generations.
Set resolution constraints via natural language. Write output resolution 1024x1024 within the prompt if the API configuration lacks explicit dimension fields for this model version.
Include negative constraints to clean up artifacts. Add no text overlays, no watermarks, clean background to prevent common rendering errors that require post-processing.
Chain requests for variations using seeds. Send five identical prompts with different seed values to select the best composition programmatically without manual intervention.
Validate output format strictly. Check the response JSON for inlineData containing the base64 image string before attempting decoding to avoid crashing your script on empty responses.

Common mistakes

Hardcoding credentials in scripts: Developers often paste keys directly into source files for speed. Fix this by using a .env file loaded via a library like python-dotenv to keep secrets out of version control systems permanently.
Ignoring rate limit headers: Free tiers reject requests silently after hitting thresholds during testing. Fix this by implementing exponential backoff logic in your retry mechanism when receiving 429 status codes from the server.
Overloading single prompts: Asking for complex scenes with multiple subjects in one go fails often. Fix this by breaking complex imagery into separate generation calls and compositing them later using standard image editing tools.

Try this in MidassAI

For a managed experience without handling raw API keys, use the MidassAI interface directly. It wraps the Nano Banana 2 capabilities into a visual workflow builder, allowing you to test prompts and save iterations without writing boilerplate code or managing security credentials. Start building your image generation pipeline directly at https://www.midassai.com/studio/nano/ to streamline production deployment and integrate with your existing marketing tools.

Nano Banana 2 in AI Studio: Free Tier to Your First API Image