Nano Banana 2.1 halves AI image prices: check the input tokens

Google released Nano Banana 2.1 on October 6, and the headline number is real: a standard 1K image from the Gemini API now costs $0.0336, about half of Nano Banana 2. For a store that generates lifestyle shots, ad variants, or catalog images at volume, that is a big cut. But one line on the pricing page moved the other way, and if your pipeline feeds the model a lot of reference photos, you need to do the math before you switch.

What actually happened

Per Google's Gemini API pricing page:

• Image output dropped by half. Nano Banana 2.1 (gemini-nano-banana-2.1) bills image output at $30 per million tokens, down from $60 on Nano Banana 2. That works out to $0.0336 per 1K image, $0.0504 per 2K, and $0.113 per 4K. Nano Banana 2 was $0.067, $0.101, and $0.151.
• Batch halves it again. Through the Batch API, a 1K image is $0.0168.
• Input went up. Text, image, and video input is now $1.50 per million tokens, up from $0.50. Text and thinking output went from $3.00 to $7.50 per million.
• Where it runs. Google's model card lists the Gemini app, AI Studio, the Gemini API, AI Mode in Search, Google Ads, Flow, and Stitch.

Heise reports the model takes up to 14 reference images and improves text rendering and character consistency, the two things that usually wreck AI product shots.

Why it matters for your business

For most small brands, the image is the expensive part, so 2.1 is cheaper. A run of 1,000 1K ad variants drops from about $67 to about $34, or about $17 in batch. That turns "test 40 creatives per product" from a budget line into a rounding error.

The catch is the input side. Product-photo pipelines do not send one short prompt. They send the packshot, the logo, a style reference, maybe a model photo, plus a long brand prompt. Each of those is input tokens, and input now costs 3x more. If your prompts are heavy and your outputs are small, your total bill can go up, not down.

What we would do this week:

1. Pull last month's usage. Split your spend into input tokens and image output. That ratio decides whether 2.1 saves you money.
2. Re-run 20 real jobs on 2.1. Same references, same prompts. Compare cost per finished image and how many you throw away. A cheaper image you regenerate three times is not cheaper.
3. Move overnight work to batch. Catalog refreshes and seasonal variants do not need real-time. Batch is half price.
4. Trim the references. Send the 3 photos that matter, not all 14 because you can.

Sources: Google Gemini API pricing, Google DeepMind: Nano Banana 2.1 model card, Heise: Nano Banana 2.1 lowers image prices.