Nano Banana 2 Edit (fal.ai)

Same Google model as the plain Nano Banana 2 node, but with up to four image inputs and an edit instruction. FAL_NanoBanana2Edit sends your images and your prompt to fal's fal-ai/nano-banana-2/edit endpoint and returns the edited result as a normal IMAGE. If you've ever wanted to hand a generative editor a reference photo plus a sentence - "put this coat on the person in this other photo," "redraw this scene at golden hour" - this is the node that does it without you leaving ComfyUI.

It's closed and cloud-hosted, like every Nano Banana: no weights, no local compute, per-call billing. The trade is the same one you take with the whole family - Google-grade realism and speed wrapped in a strict safety filter and an invisible SynthID watermark.

How it works

The node uploads each connected image tensor to fal and passes the list as image_urls to the edit endpoint. The model treats the first image as the primary subject and the rest as context or reference - that's the standard Google multi-image setup - then regenerates the frame per your instruction. Same shared inputs as the T2I sibling:

• image (required) plus prompt (required) - the primary photo and the edit command.
• image2 / image3 / image4 - optional reference images. The real differentiator of the edit node: multi-image jobs that a single-image editor can't do.
• aspect_ratio (extended list, default auto), resolution (0.5K–4K, default 1K), safety_tolerance (1–6, default 4), thinking_level (none/minimal/high, default none), enable_web_search (default off).
• num_images (1–4), seed (-1 random), output_format (png/jpeg/webp), private_mode.

Output is a single IMAGE tensor, ready to preview or save. A note on batches: each connected input is uploaded as the first frame of its tensor - the node isn't going to process a batch of images per input port.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/elle-trudgett/fal-ai-comfyui-nodes
pip install -r fal-ai-comfyui-nodes/requirements.txt

or search ComfyUI fal.ai Nodes in ComfyUI Manager. No model downloads - that's the point. Set your key: export FAL_KEY="...", or drop a config.ini with [API] / FAL_KEY in the node folder (the ComfyUI Desktop route). Keys at fal.ai/dashboard/keys. Restart after installing.

Gotchas

• "FAL_KEY not set" - env var or config.ini, restart, done.
• The safety filter is the boss. Google's editor will refuse instructions it doesn't like regardless of what you feed in; safety_tolerance nudges, it doesn't disable.
• More images, more uploads, more money. Every connected input is uploaded and every generation is billed; a four-image edit at 4K is not a cheap call.
• The re-emitted frame problem. Like other full-frame editors, pixels you didn't mention can shift. If you need surgical edits, mask-and-inpaint is still the tool.
• Errors surface as fal.ai error: ... with fal's own message, so read it before rerunning.

The honest pitch: use this when you want Google-grade edit quality and multi-image reference handling with zero setup. Just budget for it and remember the filter and the watermark live here too.