Comfly_nano_banana_edit

The middle child of the Nano Banana family, and the one with the honest limitation spelled out in its own README entry. ComflyNanoBananaEdit runs the original nano-banana (and its -hd sibling) with a real aspect-ratio menu and up to four reference images, in either text2img or img2img mode. It was added to the pack specifically to give the base nano-banana model a size selector - but the author is upfront that the size controls only work in text-to-image mode. Run it as img2img and the output follows your input image, size dial be damned.

If that sounds like a slightly narrower tool than ComflyNanoBanana2Edit, that's accurate - this is the lean, cheaper-feeling version of the same idea, for when you want Google-quality editing without the 14-reference circus.

How it works

Prompt + mode + up to four reference images go to the nano-banana model through the proxy; the node converts the result back to an IMAGE tensor plus the raw response. That's it - no token dials, no 2K/4K ladder, no webhook. It's the simplicity-first member of the family.

The inputs that matter

• prompt - the instruction. For editing: "change the background to x," "make the subject do y." For mixing references, tell it which image plays which role.
• mode - text2img (default) or img2img. Remember: aspect_ratio only takes effect in text2img.
• model - nano-banana (default) or nano-banana-hd. HD for extra detail, at presumably some extra cost.
• aspect_ratio - ten choices (16:9, 4:3, 4:5, 3:2, 1:1, 2:3, 3:4, 5:4, 9:16, 21:9), default 1:1. This is the whole point of the node - set your canvas in text2img mode.
• image1–image4 - reference images for editing/compositing.
• response_format - url (default) or b64_json.
• seed / webhook - lock a take / get pinged.

Outputs: image (IMAGE) and response (STRING). Note: no image_url output on this one - the tensor is your copy, so save it or you lose it.

Install

Manager → Comfyui_Comfly, or:

cd ComfyUI/custom_nodes
git clone https://github.com/ainewsto/Comfyui_Comfly

Restart. No model downloads; the requirements.txt is the standard aiohttp/requests/Pillow/numpy set. Key from ai.comfly.chat in Comflyapi.json or the node's apikey.

Where people get burned

The mode/size trap is the headline: set a 9:16 aspect ratio, run it in img2img, and get a square-ish output - not a bug, a documented quirk, so pick text2img when canvas size matters. And since there's no image_url output, an unsaved result is gone once the node's done; if you want the hosted copy for later, use ComflyNanoBanana or the 2-edit node instead. Otherwise this is the "just edit the picture" node - keep it simple and it behaves.