SF WaveSpeed Nano Banana Text to Image

Nano Banana is Google's Gemini-based image model - the one that does clean text-in-image, good style transfer, and generally behaves like a frontier generator without the local-VRAM tax. SF WaveSpeed Nano Banana Text to Image is the no-frills text-to-image wrapper for it: prompt in, image out, nothing clever. When you don't need multi-reference editing or pairs of candidates, this is the node you reach for first.

It's part of the comfyui-stillfront pack, and like all the WaveSpeed nodes it's a paid API call - no model weights, no GPU requirements, just your WaveSpeed key and a prompt.

How it works

The node sends your prompt to WaveSpeed's Nano Banana endpoint, waits for the generation, downloads the resulting image, and hands it back as an IMAGE tensor. That's the whole pipeline. The simplicity is the selling point: one call, one image, ready to drop into a preview or save node.

The inputs that matter:

• prompt - your text prompt. Nano Banana is one of the models that rewards plain, direct language over prompt-engineering gymnastics.
• seed - default -1 (random); set a fixed value for reproducible output. It has a "control after generate" widget so it re-randomizes each run unless you lock it.
• output_format - jpeg, png, or webp. PNG if you need transparency, jpeg for photo-ish output at smaller size.
• enable_sync_mode - default on; the node waits for the result either way and returns the image.

Output is a single output_image (IMAGE).

Installing it

Pack install, then a WaveSpeed key:

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

Restart ComfyUI. Then add your key to the SF WaveSpeed Client node (which wires into this node's client input), put it in config.ini (copy from config.ini.tmp), or set the WAVESPEED_API_KEY environment variable.

Troubleshooting

• "Unauthorized" or "API key is empty." The client node isn't getting a key. Check the api_key field, then config.ini, then the env var - it checks in that order.
• Output format choices don't include the one I want. jpeg/png/webp is all WaveSpeed's API offers here. For higher-res output you're looking at the wrong node; Nano Banana's 1K native resolution is a fixed ceiling on this endpoint.
• Image quality varies run to run. That's the seed being random (and Nano Banana being a "powerful but unimaginative" model when it's nervous about your prompt - Google's filters are in the loop even via WaveSpeed).
• Cost. Each image is a billable generation. Nothing exotic, but it's not free, and Nano Banana's hosted price is higher than the consumer-grade models - fine for occasional use, worth watching on a batch.

Honestly, this is the "just works" entry point for Nano Banana in this pack. If you want the whole Pro feature set - 4K output, reference images, thinking mode - the Pro siblings in the pack are where you go. This one is for when you want the model's quality with the least ceremony.