Nano Banana Multi Input - Klinter

"Nano Banana" is what the community calls Google's Gemini image model (the gemini-2.5-flash-image endpoint), and it's a genuinely impressive thing - real instruction-following image generation and editing, decent aspect-ratio control, all through an API. Nano Banana Multi Input - Klinter is a ComfyUI front end for that API with a twist: it takes up to five input images, so you can do multi-image composition and editing right inside a node graph instead of wrestling with a browser.

The name is a bit of a lie in one sense - despite the yellow banana theming, this node calls a paid cloud API and needs a key. It reads GEMINI_API_KEY or GOOGLE_API_KEY from your environment, and there's an api_key input to override it per-node (leave blank to use the environment). Without a key you get warnings and error paths, not images.

How it works: you give it a prompt plus any number of images, it sends them to Gemini, and returns generated_image plus a description string (Gemini often returns text alongside the image). Give it images and it runs in edit/compose mode; give it none and it runs pure text-to-image. The aspect_ratio input maps to Gemini's native ImageConfig aspect ratio - pick from 1:1 up to 21:9, or passthrough to let the API decide. That's a genuinely nice touch, since aspect control is one of the fiddliest parts of API image generation.

Now the honest part, because you should go in clear-eyed. First, there's a real dependency step the README skips: it needs pip install google-genai, and without it the node falls back to a text-only mode that returns a description and a placeholder instead of an image. Second, it's an API call per run - every queue execution burns tokens and takes seconds, and it's not free. Third, the codebase has rough edges (an initialized attribute is referenced but never set, so some fallback paths behave oddly). It works, but this is a one-person pack, not a polished commercial integration. Treat failures as "check your key, check the library, check your quota" rather than panic.

Install: search "Klinter_nodes" in ComfyUI Manager, or cd ComfyUI/custom_nodes && git clone https://github.com/klinter007/klinter_nodes, then pip install google-genai, set your key, and restart. No model files to download - it's all remote.

If you already pay for a Gemini API key and want multi-image editing inside ComfyUI, this is a convenient bridge. If you don't have a key, skip it - local models are the better default anyway.