🎨 Nano Banana

Google's Nano Banana line is actually three different models wearing one meme name, and this node is the pack's unified access point for all of them - pick the tier from a dropdown instead of wiring up three separate nodes. It's an API call, not a local model: no GPU load, no download, just your Gemini key and Google's servers doing the work.

The three models behind the dropdown

The "Nano Banana" branding started as an anonymous codename that leaked on a HuggingFace leaderboard; Google's own people started dropping banana emojis about it on social media, and the name stuck hard enough to become official. Underneath it:

• Nano Banana - Gemini 2.5 Flash Image, the base/consumer tier. Fast, cheap, good enough for most everyday generation and editing.
• Nano Banana Pro - Gemini 3 Pro Image, the flagship. Native 4K output, up to 14 reference images, and a "thinking mode" that reasons about the prompt before drawing.
• Nano Banana 2 - Gemini 3.1 Flash, a February 2026 release that's the default in this node. A speed/quality hybrid, positioned between the other two.

If you've used the pack's separate Gemini Flash 2 Image or Gemini 3 Pro Image nodes, this node is the same underlying API surface collapsed into one place - pick the tier per-run instead of building three parallel branches for it.

The inputs and outputs that matter

• Gemini_API - your key, or leave blank to pull from the pack's API_key.json.
• model - the tier: Nano Banana, Nano Banana Pro, or Nano Banana 2 (default).
• prompt / sub_prompt - the main instruction plus a second box for extra detail.
• image (optional) - a reference image to edit or compose from.
• aspect_ratio - Auto plus fifteen presets total (square, portrait, landscape, ultrawide and more). Pick one deliberately if framing matters - Auto is a guess.
• resolution - roughly 0.5K through 4K in four steps. This is the cost dial across all three tiers; higher resolution costs more per image, so draft low and finalize high rather than the reverse.
• translate - auto-translates prompt text into English from over a hundred languages before sending it.
• seed - for reproducibility where the underlying model respects it.

Three outputs: image, text_output (any accompanying text the model returns - Pro's "thinking mode" in particular tends to explain itself), and prompt (the final prompt actually sent, post-translation).

Installing it

ComfyUI Manager: search SDVN_Comfy_node, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node

Then pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt from your ComfyUI root, and restart. Get a Gemini key at aistudio.google.com/app/apikey, and put it in API_key.json (renamed from API_key.json.example) rather than pasting it into the node - that way it doesn't ride along inside a saved workflow you share.

Common issues & troubleshooting

Switching tiers changes both cost and quality, so don't leave it on the expensive default for drafts. Nano Banana 2 is the node's default, but for quick iteration the cheaper base Nano Banana tier is usually the better call - save Pro's 4K/14-reference muscle for the shot you're actually finalizing.

Refusals and restricted subjects. Google's safety filtering is strict and got noticeably stricter in early 2026, particularly around real people and known IP. A refused prompt just comes back empty - that's server-side, and nothing in the node can override it.

Every image carries an invisible SynthID watermark. Fine for most uses, worth knowing if provenance matters for what you're making.

Empty output with a key that looks valid. Usually a billing or quota issue on the Google account behind the key, not a bug in the node - check the key's status in AI Studio before assuming something's wrong on this end.