WaveSpeedAI Google Nano Banana Pro T2I

Nano Banana Pro is the model people describe as "the one that finally got it" - Gemini 3 Pro Image, 4K-native output, multilingual text rendering, and the kind of prompt understanding that makes a one-line description come out fully realized. It's closed, it's Google's, and normally you reach it through ImageFX or the Gemini API. This node hands you the same model from inside ComfyUI, billed per image by WaveSpeed AI. For text-to-image specifically, it's the closest thing in this pack to "the model does the work."

What it is

The text-to-image member of the pack's Google trio. prompt is the entire job; aspect_ratio offers ten options (1:1 through 21:9); resolution runs 1k / 2k / 4k at the tooltip's prices ($0.14 / $0.14 / $0.24). No reference images, no LoRAs, no negatives - this is a pure generation node, and its simplicity is a feature. There's no optional input section at all; the info schema is just client, prompt, ratio, resolution, format, and sync mode.

How it works

client from the WaveSpeedAI Client node, then a POST to the Nano Banana text-to-image endpoint with the prompt, aspect ratio, and resolution. The output downloads back as an IMAGE tensor. output_format defaults to png for transparency support (Nano Banana can generate real alpha), and enable_sync_mode defaults on, so the node blocks until the image lands in your graph.

The inputs that matter

• prompt - describe the image; this model rewards description over keyword soup. It reasons about the prompt before generating, so prose works.
• aspect_ratio - pick the shape up front; Nano Banana's output is native to it.
• resolution - 1k/2k/4k. 2k costs the same as 1k, so there's little reason to stay low unless you want quick drafts.
• output_format - png (transparency) or jpeg.

Output name is output_image, type IMAGE - wire into Save Image / Preview Image.

Install & setup

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/razv-wavespeed.git
cd razv-wavespeed
pip install -r requirements.txt

or ComfyUI Manager → "razv-wavespeed" → Install, restart. API key from wavespeed.ai in config.ini (from config.ini.tmp), the Client node, or WAVESPEED_API_KEY. Deps: requests, pillow, numpy<2.0.0, torch.

Common issues

• Blank or refused generations - Nano Banana's safety filter is aggressive. If a prompt trips it, you get an error or a refusal, not a warning; rephrase rather than retry.
• Text rendering off - spell what you want in quotes; Pro handles multilingual text well but only what you give it.
• "No output received" - transient API error or timeout; retry.
• 401 - key issue.

The honest comparison: for pure T2I you have real open alternatives locally that cost nothing per image, and Nano Banana Pro's censorship makes it a poor fit for half the reasons people generate images at all. Where this node wins is quality on demand - a hero image, a character concept, a specific render where you want Gemini's understanding without Gemini's interface. When you need it, it's the most plug-and-play node in the pack.