If your team makes ad creative or social images through the Gemini API, this is a maintenance task rather than a strategy shift. The model you probably use was deprecated on October 6, and its replacement is cheaper per image at common sizes but charges more for input. Here’s the short version for Alberta marketing operations.
What Google released
“An update to Nano Banana 2 (Gemini 3.1 Flash Image), it maintains Flash-level speed and cost efficiency while delivering significant improvements in visual quality, prompt adherence, multi-turn character consistency, and text rendering.”Google AI for Developers, Gemini Nano Banana 2.1 model page, updated October 6, 2026
Google’s model page lists 1K, 2K and 4K output (1K by default), fixes for tiling artifacts on wide and panoramic ratios, up to 14 reference images, Google Search grounding and Batch API support. Google DeepMind’s model card says it is based on Gemini 3.6 Flash. It lists the model across the Gemini app, AI Studio, the Gemini API, Google Ads and other Google tools. The card also lists known limits, including blurry small text at 1K and imperfect character consistency.
What to change
Swap the model ID from gemini-3.1-flash-image to gemini-nano-banana-2.1 in your scripts, automations and any no-code tools that call the API. Google’s deprecations page shows no shutdown date yet, so there’s time to test. Run your last month’s best-performing prompts through both models, check text-heavy layouts and brand colours by eye, and confirm aspect ratios. Nano Banana 2 offers a 0.5K size on the pricing page, but the 2.1 model page lists only 1K, 2K and 4K. If you use 0.5K thumbnails, plan for that.
Cost check
| Item | Nano Banana 2.1 | Nano Banana 2 (gemini-3.1-flash-image) |
|---|---|---|
| Input, per 1M tokens | $1.50 | $0.50 |
| Text and thinking output, per 1M tokens | $7.50 | $3 |
| Image at 1K | $0.0336 | $0.067 |
| Image at 2K | $0.0504 | $0.101 |
| Image at 4K | $0.113 | $0.151 |
From these listed rates, 100 images at 1K come to $3.36 on 2.1 and $6.70 on Nano Banana 2, before input and thinking tokens. That is our arithmetic, not a measured bill. Workflows that send many reference images or long prompts pay three times the input rate on 2.1, so check a week of real usage before assuming savings. Prices are in US dollars; confirm currency and taxes on your account.
Labelling AI images (our advice)
Google’s image generation docs say all generated images include a SynthID watermark. That watermark is invisible to viewers, so it doesn’t replace a visible label. Our advice is to follow each ad platform’s AI-content rules, add a short “AI-generated” note where your audience or client would expect one, and never use AI images to show results, products or job sites that aren’t real. Keep a simple log of the prompt, model ID and date for each published image. Our AI social content page covers how we run social creative for Alberta businesses.