NS WaveSpeed Nano Banana Text to Image

Nano Banana - the model, not the whole Pro line - is Google's Gemini 2.5 Flash Image behind the goofy name that stuck. It's the consumer-grade, speed-first member of the family, and this node is its text-to-image mode over WaveSpeed's API. If you've only heard the Pro hype, this is the cheaper, faster, seed-controllable version.

It's part of the Symbiotica pack's neuralsins/WaveSpeed family. What separates this node from its Pro siblings is exactly the two things that matter to a tinkerer: it exposes a seed, and it's the budget end of the lineup. No open weights exist for any of these, so "inside ComfyUI" only happens through an API - this is one of the lighter ways in.

How it works

prompt in, output_image (IMAGE) out, via WaveSpeed's /api/v3/google/nano-banana/text-to-image. The inputs:

• prompt - the image description. Base Nano Banana is fast and capable, though it won't hold a candle to Pro on a complex multi-part instruction.
• seed - -1 for random, otherwise a fixed seed for reproducible draws. This is the input the Pro nodes conspicuously lack.
• output_format - jpeg (default), png, or webp. The three-way menu is slightly broader than the Pro nodes' two.
• enable_sync_mode - on by default; the node waits and returns the finished tensor.

Note what's absent: no aspect_ratio, no resolution, no reference images. This node is deliberately bare - it generates at the model's default size. The edit variant (NS WaveSpeed Nano Banana Edit) handles image inputs, and the Pro line handles the size ladder.

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/symbiotica-ai/comfyui-nodes.git symbiotica
pip install -r symbiotica/requirements.txt

Restart, or search Symbiotica in ComfyUI Manager. Add an NS WaveSpeed Client node with your WAVESPEED_API_KEY (Settings → Symbiotica, or the env var) and wire its client output in.

Common issues

• "Reproducible" with -1 means nothing. Leave the seed at -1 and every run is a fresh draw. Set a fixed seed only when you want the same result back.
• No size control. If you need a specific aspect ratio or 4K, this isn't the node - the Pro Text to Image variant is.
• Every run is a paid call. It's the cheap end of the family, but it's still metered. Mute the branch when you're happy.
• The model's filter is baked in. Base Nano Banana's safety behavior is the most complained-about of Google's image models - the KB's closed-source doc quotes ImageFX criticism of its "hidden biasing." What it refuses, it refuses; no node setting changes that.

The honest take: this is the "quick draft" node of the Nano Banana family - fast, cheap, seedable, and deliberately un-fancy. When the draft gets serious, the Pro line is one node swap away.