SF WaveSpeed Nano Banana Pro Text to Image Ultra

This is the biggest canvas Nano Banana Pro will sell you. SF WaveSpeed Nano Banana Pro Text to Image Ultra is the standard text-to-image node with the resolution dial extended to 8K - 4k at $0.15, 8k at $0.18, straight from the tooltip - and 8k as the default. If you've ever needed a poster-sized image straight out of the generator instead of upscaling a 1K render and hoping, this is the node. The closed Gemini model does the thinking, WaveSpeed does the hosting, your key pays for it, and a genuinely huge IMAGE tensor comes back into your graph.

How it works

Same shape as the standard Nano Banana Pro text-to-image node - prompt, aspect ratio, resolution, output format, sync mode - but the endpoint is google/nano-banana-pro/text-to-image-ultra and the resolution enum is 4k / 8k. The node posts and returns a single IMAGE. Mechanically there's nothing exotic; the interesting part is what you do with an 8K tensor after it lands. If you're planning to save or composite it, make sure the downstream nodes and your save format can handle the size, or the graph will quietly become the slowest part of your day.

The inputs that matter

• client - from SF WaveSpeed Client, wired first.
• prompt - the generation prompt. At 8K there's more surface area for mistakes, so composition language matters more, not less.
• aspect_ratio - default 1:1.
• resolution - 4k ($0.15) or 8k ($0.18), 8k default.
• output_format - PNG (default) or JPEG.
• enable_sync_mode - default true.

Output: output_image (IMAGE).

Installing it

In the SF ComfyUI Nodes pack from Stillfront. ComfyUI Manager → "SF ComfyUI Nodes" → Install → restart, or:

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

Plus a WaveSpeed API key through SF WaveSpeed Client.

Common issues

The honest advice for this node: resist the default. 8K is impressive but it's a bigger, slower call for a marginal quality gain over 4K on most subjects - the KB's caveat about the Nano Banana line being powerful yet filter-limited applies here too. Use 4k for anything that isn't a print or a billboard, and save the 8k for the actual deliverable. And if a big tensor makes your downstream preview nodes choke, that's not the generation failing - split the pipeline so Ultra outputs straight to a save node. Everything else is the same family script: it's closed, it's filtered, and it costs real money per call, so iterate small and go big once.