google / nano-banana-2-lite

Nano Banana is the closed model with the weirdest name and the biggest demand in the ecosystem - Google's Gemini-native image generation, so heavily requested inside ComfyUI that it saturates community discussion. You cannot download it, ever; the only way in is an API. This node is the Nano Banana 2 Lite variant running through Civitai's orchestration layer. Prompt in, IMAGE out, billed in Buzz.

"Lite" is the speed/cost end of the Nano Banana 2 line - Google's "speed-quality hybrid" family, with the lite variant aimed at cheap, fast output. It's the pragmatic pick when you want the Gemini-image look for drafts or bulk without paying flagship Nano Banana Pro prices. The KB's honest summary of the family: technically impressive, heavily filtered ("powerful, unimaginative"), with invisible SynthID watermarks baked into every output.

How it works

Standard civitai-comfy-nodes mechanics: the node submits a nano-banana-2-lite imageGen job to the Civitai Orchestration API, long-polls, and converts the result to a native IMAGE tensor. Nothing runs locally. Notably, the node also takes an images input even though it's a create node - that's the reference-image path Nano Banana supports (edit/reference style), so this one node covers both plain generation and reference-based generation.

The inputs that matter

• prompt - required. Nano Banana rewards detailed natural language.
• images - optional reference images (a prior IMAGE, or Load Image). Wire them for style/identity references.
• aspect_ratio - from 21:9 and 16:9 down through 1:1 (default) to 9:16 portrait. No auto here, unlike some sibling nodes - pick one.
• num_images - 1 to 4 per job (note: this node calls it num_images, not quantity).
• seed - present, unlike the fal nodes, so you can reproduce results.
• output_format / image_metadata - format and optional metadata.

Outputs

images → Preview/Save Image. errors, workflow_id, raw_json are diagnostics.

Installing and authenticating

Pack-wide install - Nano Banana nodes ship in civitai-comfy-nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

or ComfyUI Manager → "Civitai Comfy Nodes" → Install → restart. Dependencies: requests + python-socketio[client].

Auth: Civitai Auth node → CIVITAI_API_TOKEN → stored key/OAuth → browser login. Headless:

export CIVITAI_API_TOKEN=your_token

Where people get burned

• The filter is the model. Nano Banana's IMAGE_SAFETY filtering is aggressive and got stricter in 2026, including limits on famous IP and celebrity likeness. There is no bypass - no weights to edit. If your work lives near that boundary, this is the wrong tool and it's not the node's fault.
• SynthID watermark. Every output carries Google's invisible watermark. Fine for most use, a blocker if you need provably clean output.
• Billed per image. This is the "cheap" Nano Banana, but cheap is relative - a batch of num_images: 4 is four billed calls.
• It's the lite variant. Expect the speed/cost tradeoff in quality. If you want the 4K flagship look, Nano Banana Pro (a different node in this pack) is the upgrade path.
• Your prompt and reference images go to Google via Civitai's orchestration. Standard API-node deal.