Nano Banana Pro Edit (fal.ai)

The edit half of the Nano Banana Pro pair. Where FAL_NanoBananaPro generates from a prompt alone, FAL_NanoBananaProEdit takes up to four images plus an instruction and sends them to fal's fal-ai/nano-banana-pro/edit endpoint. It's Google's flagship Gemini 3 Pro Image doing the work, which means 4K-class output, strong realism, and the usual Google caveats: closed, cloud-hosted, per-call billing, a strict safety filter, and an invisible SynthID watermark on everything.

If you want to say "here are four photos, now give me a scene that uses this character, this outfit, and this lighting" without stitching a Frankenstein workflow of open models together, this node is the clean version of that ask.

How it works

Every connected image tensor is uploaded to fal and passed as image_urls; the first is the primary, the rest serve as reference. Your prompt drives the edit, the result comes back as a standard IMAGE tensor, ready to preview or save.

The inputs that matter:

• image (required) and prompt (required) - the main photo and the edit command.
• image2 / image3 / image4 - reference images, the reason to pick this over a single-image editor.
• resolution - 1K, 2K, 4K; default 1K. Again the dominant cost lever.
• aspect_ratio - standard set (1:1 default, 4:3, 16:9, 3:2...), not the extended ultrawide list.
• safety_tolerance (1–6, default 4) - a filter nudge, not a bypass.
• enable_web_search (off by default), num_images (1–4), seed (-1 random), output_format (png/jpeg/webp), private_mode.

One IMAGE output. Each input port takes the first frame of whatever tensor you connect, so this isn't a batch-processor - one image per 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 ComfyUI Manager → ComfyUI fal.ai Nodes. No model downloads - the model never leaves Google's servers. Set your key: export FAL_KEY="...", or a config.ini with [API] / FAL_KEY in the node folder for ComfyUI Desktop. Keys at fal.ai/dashboard/keys, then restart.

Gotchas

• "FAL_KEY not set" - env var or config.ini, restart.
• This is the most expensive node in the pack. Four uploaded images plus a Pro-tier edit generation per run. Reserve 4K for the final pass, prototype at 1K.
• The filter is baked in. Instructions Google's stack dislikes get refused no matter what safety_tolerance says, and the output carries the watermark.
• Full-frame re-emission. The model redraws the whole image, so details you didn't ask to change can drift. Precision jobs still want masking.
• Errors come back as fal.ai error: ... with the API's message intact.

It's a fun node for character-consistency and multi-reference shots, and honestly the cleanest way to get Google's flagship editing quality in ComfyUI. Just treat it like the premium it is: fewer, more deliberate calls.