Nano Banana 2.1: Image API Pricing, Edits and Prompts

These Gemini Developer API rates were checked on October 7, 2026. The table covers image output alone. Batch halves the listed output charge; input and text/thinking have separate rates. Batch jobs are asynchronous and can take up to 24 hours; use Standard for interactive edits. There is no free API tier for this model.

A new output on an edit turn is another generated image. For 100 finished 2K assets, an illustrative budget of one initial generation and two revisions per asset means 300 outputs, or $15.12 in Standard image charges. If each first attempt is accepted, the same 100 assets cost $5.04 in image charges.

Standard input costs $1.50 per million tokens; output text and thinking cost $7.50 per million. Batch rates are $0.75 and $3.75 respectively. Record usage across all turns, including reference images and conversation context, rather than multiplying only the final asset count.

Google Web and Image Search share a 5,000-search monthly allowance across Gemini 3 and newer models. Beyond it, searches cost $14 per 1,000. A generation can trigger multiple searches; retrieved grounding context is not charged as input. Ground factual imagery when needed and include search usage in the job budget.

Open the model in Google AI Studio to try a prompt, or use a Gemini API project with billing enabled. For an API request, keep the key in your environment and select the stable model ID. The request below asks for a 2K image with a square aspect ratio; its image-output charge is $0.0504 before other usage.

The response contains the generated image data. Save that output, inspect the text and composition, and keep the interaction ID if you want to continue editing. Google’s guide shows how to decode and save an image in Python or JavaScript.

Google’s Nano Banana 2.1 showcase includes text-rich posters, a masked background edit and a handbag campaign. Open the new 2.1 section to inspect those outputs and their original prompts. Use the HowMuch starter prompts below for similar tasks. The following examples link their authors’ prompts and outputs.

For a poster, separate exact copy from layout instructions: “Create a vertical poster for a fictional ceramics workshop. Use only the words CLAY DAY, SUNDAY 12–5 and ALL LEVELS. Put one terracotta bowl below the title, cream background, navy lettering, generous margins.” Check every letter before accepting it.

For a reference-led product variation: “Use the supplied mug photo. Keep the handle shape, glaze, logo and camera angle. Replace only the background with a warm kitchen counter in morning light. Do not add text.” Compare the handle and logo with the input; count revisions as additional outputs.

For a targeted edit: “Keep the supplied subject at the same size and position. Replace the background with a rainy garden and match the lighting. Preserve the subject’s outline and surface details.” Keep the input beside the result when checking what changed.

Fuser’s October 6 demonstration identifies Nano Banana 2.1 at 1K. Its product brief specifies a speckled ceramic dripper, a pale stone plinth, warm plaster and light from the left. One edit swaps the plinth for walnut while preserving the product and setting. A separate branch combines the original stone-base image with a logo reference.

The original comparison makes the branch visible: the logo version retains the stone base. For your own prompt, name both the intended change and the details to retain. Compare the rim, handle, glaze and shadow as well as the edited area.

As an illustrative Gemini API budget, three 1K outputs are $0.1008 in image charges. Logo generation, reference inputs, reasoning and retries add usage. This example counts three outputs; Fuser does not report total attempts or billed usage.

In an October 6 X thread, ibexdream demonstrates Nano Banana 2.1 through Krea. ibexdream’s profile identifies them as a Krea creative partner. The workflow starts with an illustrated banknote, edits its lettering and symbols, then tries a different visual style. The linked result is the text edit, not the later style variation.

The output contains the requested central slogan and serial-number motifs, while the corner ornaments merit a closer check. For this kind of edit, list each text replacement and its location; inspect unchanged decoration too. A readable headline alone does not show that every constraint survived.

Krea’s charge and total attempts are not stated in the post. Keep the platform’s billing separate from the Gemini API rate table above, and count each retained or rejected edit when comparing workflow cost.

Zac shares a long Chinese landscape scroll identified as Nano Banana 2.1 output. The author reports a first-pass 2928 × 352 result and points out a lettering failure. The author’s public preview is only 680 × 82, so it is useful for the composition, not for independently checking every character.

For a text-bearing panorama, inspect the original file at full size before accepting it. Keep a separate text reference and check line breaks, omissions and duplicated characters. Do not map the preview dimensions to an API resolution tier or infer a cost from them; this post does not provide a bill.

Start at the default 1K for drafts; choose 2K or 4K when the delivery size or fine detail calls for it. Nano Banana 2.1 does not support the old model’s 512px output. Google reports improvements to wide and panoramic compositions at 2K and 4K, including 1:4, 4:1, 1:8 and 8:1 aspect ratios.

For reference-led work, describe what each image contributes: the product, person, pose, background or layout. The model supports up to 14 references, with fidelity guidance covering up to 10 objects and four characters. More references also increase the material the model must process.

Thinking defaults to medium; minimal and high are also available. Compare settings on the same brief and inspect the resulting usage. A higher setting changes the reasoning budget, so the per-image output price alone cannot predict the whole request cost.

Google released Nano Banana 2.1 on October 6, 2026. The previous stable model, gemini-3.1-flash-image, is deprecated, with no announced shutdown date in Google’s current schedule. Replace that model ID with gemini-nano-banana-2.1, then check resolution, thinking settings and a representative edit sequence before switching production traffic.

The image-token rate falls from $60 to $30 per million, but output token counts also change with resolution. At 4K, the old model uses 2,520 image tokens ($0.1512), while 2.1 uses 3,780 ($0.1134): a 25% reduction in image-output cost. At 1K and 2K, image-output cost halves. Input and text/thinking rates rise. Reprice the full workflow rather than assuming the total bill halves. Also replace any 512px setting with a supported size; the new default thinking level is medium instead of minimal.

Start with 2.1 for product variations and repeated poster edits. Google positions it for speed and volume, and Pro for complex design control. If 2.1 misses a layout or brand constraint, test Pro with the same prompt, references and resolution. Check line breaks, logo shape and color, and areas that should stay unchanged; a higher price alone does not settle the choice.

The stable API IDs are gemini-nano-banana-2.1 and gemini-3-pro-image. These Gemini Developer API prices, checked October 7, 2026, cover image output only; include input, text/thinking, search and any billed retries in the task budget. Use Standard for interactive work and asynchronous Batch when turnaround allows.

For 100 outputs at 2K, Standard image charges are $5.04 with 2.1 and $13.44 with Pro. Compare the total spend and edits needed to obtain accepted assets on your own brief.

Google positions Nano Banana 2 Lite for high-volume 1K generation. Compare it when 1K output meets the brief and you want another hosted option for a large batch.

If local inference or downloadable weights are part of the requirement, compare an open-weight image workflow. Nano Banana 2.1 is an API service: installing a client does not make the model run on your GPU.