WaveSpeedAI Google Nano Banana Text-to-Image

If you're going to use this pack for images at all, this is probably the node you use most. It's the plain text-to-image version of Google's Gemini 2.5 Flash Image - the model the community renamed "Nano Banana" after it took over every image feed in late 2025 - served through the WaveSpeed AI API by razvanmatei-sf/razv-wavespeed. No Pro tier, no Ultra resolution, no batch-of-two gimmick. Just prompt in, image out, and it's quietly the best value-per-image in the whole pack.

What Nano Banana is good at matters here: it renders text and logos that stay legible, follows style descriptions well, and handles natural-light photography without that AI-wax look. If you've been fighting local models for a "menu with the right words on it" and losing, this is the model that ends the war.

How it works

Wire a WaveSpeedAI Client node into client, type your prompt, hit go. The node POSTs to the Nano Banana text-to-image endpoint and returns an IMAGE tensor (output_image) you can preview, save, or composite. It's the least-surprising node in the pack, and that's the compliment.

The input surface is smaller than the Pro variants, and here's the thing worth noticing: there's no aspect-ratio dropdown and no size control. The model decides framing from the prompt. Want a portrait? Say "vertical composition, 2:3." Want a banner? Say "wide cinematic framing." It's a real skill adjustment if you're used to dialing in 1024×1024 everywhere.

• seed - -1 for random, or set it to lock a result. One of the few plain nodes here that exposes one.
• output_format - jpeg, png, or webp. Three options here (the Pro nodes only offer two), and png still means transparency.

The inputs that matter

• prompt - your description, and with no size dial, composition lives here.
• output_format - jpeg for small files, png for transparency, webp as the modern middle ground.
• seed - set it when you find a result you want to iterate on.

Install

Shared pack:

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/razv-wavespeed
# restart ComfyUI

Or ComfyUI Manager → search "razv-wavespeed" → Install. Add a WaveSpeed AI key through the WaveSpeedAI Client node, config.ini (copy from config.ini.tmp, [API]), or a WAVESPEED_API_KEY env var.

Common issues

• No size control. The "why is my image landscape?" confusion is the #1 question here. It's not a bug - describe the framing you want.
• Paid API, popular model. Queue times spike when Nano Banana is trending, which is often. Sync mode (default) will hold the graph; flip it off for batch runs.
• The tooltip's "Upload Image node" is a red herring for this one - it's text-to-image, no upload needed.
• Missing nodes after install: if the whole WaveSpeed menu is absent, check that av, torchaudio, pydantic, and comfy_api are importable - the pack's thin requirements.txt doesn't list them but the code needs them.