Nano Banana Edit (fal)

Nano Banana is Google's Gemini image model, and it is the one most people reach for when they want to edit an image by just describing the change - "put the person from image 1 into the scene of image 2," "remove the background," "make it a rainy night." NanoBananaEdit brings that into ComfyUI: give it one or more images and a text instruction, and it returns the edited result as a native IMAGE. No masks, no inpainting fiddliness, and it's genuinely good at multi-image composition.

It runs through fal in gokayfem's ComfyUI-fal-API pack - closed model, hosted, billed per image, nothing local. This is one of the standout nodes in the whole pack, because instruction-based editing and reference compositing are exactly what Nano Banana is best at.

How it works

You supply at least one image and a prompt describing the edit; Nano Banana reasons about the instruction and the reference(s) and produces a new image. Because it's multimodal and reference-aware, you can feed it several images and ask it to combine them - a subject from one, a setting from another. fal runs it remotely and returns a native ComfyUI IMAGE, so it wires straight into a Preview or Save node.

The inputs and outputs that matter

• prompt (required) - the edit instruction. Be direct: describe the change, not the whole image.
• image_1 (required, IMAGE) - your primary image.
• image_2 / image_3 / image_4 (optional) - additional references for compositing. This is where the magic is: hand it a person and a background and tell it to merge them.
• num_images (1–4) - how many results to generate; ask for a few and pick the best edit.

There's also an images batch input and output_format (png / jpeg). The single output is an IMAGE.

How to install it

Via ComfyUI Manager: search ComfyUI-fal-API, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/gokayfem/ComfyUI-fal-API.git
pip install -r ComfyUI-fal-API/requirements.txt

No models download. Add a fal API key to config.ini under [API] (from config.ini.example) or export FAL_KEY=your_key, restart, and find it under FAL.

Common issues & troubleshooting

• Auth error. FAL_KEY not set. Fix and restart.
• The edit went too far / changed the wrong thing. Tighten the instruction. "Change only the sky to sunset, keep everything else identical" beats "make it sunset." Generate a few with num_images and pick.
• Compositing didn't pull the right element. Reference the images explicitly in the prompt ("the woman from the first image," "the room from the second") so it knows which is which.
• A prompt gets refused. Google's content filtering runs server-side; edits involving recognizable people can trip it. The pack raises fal's real error text, so you'll see the reason.
• Cost. Each generation is a paid API call, and num_images > 1 multiplies it. The result cache serves repeat identical calls free (force_rerun bypasses); the spend guard caps a session.
• Windows Portable. Missing fal-client? python_embeded\python.exe -m pip install fal-client.

If you do any kind of "take this, change that" image work, this is one of the most useful nodes in the pack. Precise instructions and clear reference images are the whole game - feed it those and Nano Banana's editing is hard to beat.