NS WaveSpeed Nano Banana Pro Text to Image

Sometimes you want Nano Banana Pro - the flagship Google image model with its reasoning step and its 4K native output - but the only thing you have is a prompt. This node is that: Pro's text-to-image mode, over WaveSpeed's API, dropping a finished IMAGE back into your graph.

It's part of the Symbiotica pack's neuralsins/WaveSpeed family, and it's the most "generator" of the Pro family - no image input, just text in, picture out. The KB's closed-source doc frames Pro as the heavy hitter of the Nano Banana lineup: it reasons about the prompt before rendering, handles multi-language text in images well, and its output is aggressively high-resolution. This node is the cheapest door to that inside a workflow.

How it works

prompt goes in, a prompt-encoded image comes out of output_image (IMAGE), via WaveSpeed's /api/v3/google/nano-banana-pro/text-to-image. Three inputs do the real work:

• prompt - the image description. Pro's thinking mode rewards a detailed, unambiguous sentence.
• aspect_ratio - ten choices from 1:1 to 21:9, default 1:1. Set this for the deliverable shape.
• resolution - 1k, 2k, or 4k, default 1k. The tooltip quotes the meter: 1k and 2k at $0.14, 4k at $0.24.

output_format (png/jpeg) and enable_sync_mode (on by default) round it out.

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. You need an NS WaveSpeed Client node with your key - Settings → Symbiotica, the WAVESPEED_API_KEY env var, or typed on the node - feeding the client socket.

Common issues

• No seed control here. Unlike the base Nano Banana node, this one doesn't expose a seed - so "reproducible results" aren't on the menu. If you need a seed, the plain NS WaveSpeed Nano Banana Text to Image node has one.
• 4K is tempting and pricier. The tooltip is explicit: 4k costs $0.24 against $0.14 for the smaller tiers. For a first draft, 1k is the sane spend.
• It's a paid call every run. Mute the branch when you're done iterating on the prompt - the meter doesn't pause for politeness.
• Expect the model's filter, not ComfyUI's. This is a closed Google model; whatever it refuses to draw, it refuses, and no node setting changes that.

The honest take: if you already run local generators, this node earns its keep for two jobs - text-heavy images (Pro is genuinely good at rendering legible text) and anywhere you need 4K without a tiling project. For everyday drafts, your local stack is still the cheaper reflex.