The fastest way to get started with Nano Banana 2 is to install the Python SDK and work through the official quickstart. After installing with pip install nano-banana-2, you create a Pipeline object, attach one or more transformation stages using the add_stage method, and call pipeline.run(source) with your input data source. A minimal working example that reads from a local file, applies a transformation function, and writes results to stdout can be set up in under 20 lines of Python. The quickstart in the official documentation walks through this exact setup with line-by-line explanations.
Once you have the basics working, the next step is to explore the configuration system. Nano Banana 2 uses a YAML-based configuration file to define pipeline stages, concurrency settings, buffer sizes, and plugin options. Separating configuration from code makes it easier to change behavior for different environments—development, staging, and production—without modifying application source. The CLI tool included with the Python SDK can validate a configuration file before you run it, catching common mistakes like missing required fields or type mismatches.
For teams building data ingestion pipelines that feed into a vector database such as Milvus, a recommended starting pattern is to configure Nano Banana 2 as the ingestion and pre-processing layer, with a Milvus output plugin handling the final write step. The project repository includes example configurations for this exact setup, along with sample datasets you can use to verify that the pipeline is working end to end before connecting it to real data.