Nano Banana 2 (BudgetPixel)

Nano Banana is the name Google's Gemini-native image generation goes by - it started as an anonymous codename on the HuggingFace leaderboard, the banana emojis happened, and the name stuck. Nano Banana 2 is the Gemini 3.1 Flash Image generation tier, and this node is how you call it from a ComfyUI graph. The thing to know up front: it's a closed model with no weights and no local path, so if you want it at all, you want it through an API node like this one.

What you get is the high-efficiency end of Google's image stack: solid text rendering, strong multi-reference composition, and output from 0.5K up to 4K. The billing is tiered by size - 50 credits at 0.5K, 65 at 1K (the default), 100 at 2K, 150 at 4K - so it's cheap for drafts and gets progressively pricier as you chase resolution. If you've compared Nano Banana to local checkpoints before, the usual pattern holds: the API version is fast and consistent, and the price is per image rather than per GPU-hour.

The inputs

• prompt (required) - your text description.
• reference_images - up to nine images for editing and multi-reference composition. These are normal IMAGE connections, so anything you've generated locally can feed them.
• size - 0.5K / 1K / 2K / 4K. This is the pricing lever: pick the smallest tier that looks right for the job.
• aspect_ratio - fourteen options, 1:1 default.
• num_images - 1–4.

Output is an images socket with core IMAGE tensors, straight into Preview Image or Save Image.

When to reach for it

Nano Banana 2 is the pack's middle sibling: cheaper and faster than Nano Banana Pro, more capable than the Lite. Use it when you want Google-quality images with several reference images in play - character consistency, product shots, "keep this object, change the scene" edits - without paying the Pro tier's 120-credit entry fee. The Lite at a flat 40 credits is better when all you need is a fast 1K image and don't need the size ladder at all.

One thing people get tripped up on: the resolution is a billing choice here, not just a quality choice. A 4K render costs three times a 1K render of the same image. If you're iterating on a concept, 1K is the sane default and 4K is for the final pass.

Install and issues

Install via ComfyUI Manager (search "BudgetPixel") or:

cd ComfyUI/custom_nodes
git clone https://github.com/budgetpixel-ai/comfyui-budgetpixel.git
pip install -r comfyui-budgetpixel/requirements.txt

Restart, find it under Add Node → BudgetPixel → Image, and set your API key via BUDGETPIXEL_API_KEY or a budgetpixel.json file - no key widget, by design, so your key never ends up in shared workflow JSON.

Standard pack caveats: a recent ComfyUI for core types, and interrupting a run doesn't cancel the server-side job - it keeps running and still bills. And since Nano Banana is a Google model, its content filters are baked in and un-patchable; anything the hosted model refuses stays refused, no local weights to abliterate.