Nano Banana 2.1 holds characters and infographics together

Google's image model just got better at the two things that made AI images hard to use at work: keeping a character the same, and putting accurate text on a page.

Google released Nano Banana 2.1 on October 6. The API ID is gemini-nano-banana-2.1, and Google's documentation describes it as a high-efficiency image generation and conversational editing model, an update to Nano Banana 2.

The gains are in consistency

The headline numbers come from Google's own model card, so treat them as claims until you test.

Against Nano Banana 2, Google reports:

• Multi-character consistency: 1106 Elo versus 978.
• Overall preference in thinking mode: 1050 Elo versus 990.
• Infographic factuality: a score of 0.521 versus 0.179.

The first and last are big jumps. The overall preference gain is modest. Google also lists fixed tiling artifacts on panoramic aspect ratios and better multi-turn character consistency.

Yes, every lab says its new image model is better. These numbers are at least aimed at real complaints: characters that drift between frames and infographics that get facts wrong.

What you can do with it

The API documentation lists these capabilities:

• Up to 14 reference images in one prompt.
• Character consistency for up to 4 characters and object fidelity for up to 10 objects.
• Output at 1K, 2K and 4K, with 1K the default.
• Configurable thinking levels: minimal, medium and high.
• Search grounding with Google Web and Image Search.
• Batch API support.

It does not support function calling, structured outputs, caching or code execution. If your pipeline depends on those, check before you switch.

Google says the model is available in the Gemini app, Google AI Studio, the Gemini API, Search AI Mode, Google Ads, Google Flow and Google Stitch.

The price, and the fine print

Google's pricing page lists the standard tier at $30 per million output image tokens. In per-image terms:

• 1K: $0.0336
• 2K: $0.0504
• 4K: $0.113

Batch pricing is half the standard output rate, at $15 per million output image tokens.

Google also lists limits. The model card says it renders small text and long paragraphs poorly, character consistency is not always perfect between input images and output, and its 3D reasoning and world knowledge are still limited. That matters, because small text is exactly what an infographic is made of.

What to do this week

Take the three prompts that broke your last workflow. A recurring character across six frames. An infographic with five real numbers. A product shot with four specific objects. Run them on 2.1 and on what you use now.

If the character holds and the numbers are right, you have a real upgrade. If not, Google's Elo scores did not survive your work.

Watch independent comparisons over the next few days, and keep your own test prompts handy.