Pro · 2 · 2 Lite
Google's Nano Banana family, one queue

Batch every
Nano Banana

Nano Banana is Google DeepMind's image line, and this page runs three of its checkpoints side by side: Nano Banana Pro, Nano Banana 2 and Nano Banana 2 Lite. Load a prompt list, pick the tier the job deserves, and one batch renders the set while you read results instead of typing the next prompt.

One per line — each line is a separate prompt.

0 prompts

Drag & drop or click to upload — PNG / JPEG / WEBP / GIF, up to 10MB each.

Will generate 0 images · 0 credits

Three tiers, and what each one is actually for

Half the internet files Pro and 2 as old and new. They are not. Nano Banana Pro is the reasoning branch built on Gemini 3 Pro, shipped November 2025. Nano Banana 2 is the speed branch built on Gemini 3.1 Flash, shipped February 2026 and made the default image model in the Gemini app that same day. Same generation, different assignments.

Nano Banana Pro — the reasoning tier

Google DeepMind calls Nano Banana Pro its strongest model for correct, legible text, including long passages and several languages at once. It borrows Gemini 3 Pro's reasoning to get chart labels right, holds a likeness across roughly five people in a scene, takes localised edits such as camera angle, colour grading and lighting, and can ground a render in a live Google Search lookup.

Nano Banana 2 — the working tier

Officially Gemini 3.1 Flash Image, described by Google as the speed-optimised variant. It keeps the family's text rendering and multilingual strength: likeness reported across up to five characters, fidelity across fourteen objects, tight instruction following, and access to Gemini's world knowledge. That is why it turns notes into diagrams and briefs into campaign art so readily.

Nano Banana 2 Lite — the draft tier

The Gemini 3.1 Flash-Lite branch, and the lightest thing in the Nano Banana family. Google has published little about it beyond that, so treat it as a contact sheet: burn through twenty phrasings, see which framing survives, then rerun the survivors on Pro.

The three Nano Banana tiers, measured against each other

Read this before you pick a chip. Every figure comes out of the same registry that bills the batch, so it matches what the form charges rather than what a launch post remembers: credits, output sizes, aspect-ratio coverage, reference-image support and prompt length, tier by tier.

ModelMade byCredits / imageResolutionsAspect ratiosReference imagesPrompt limit
GeminiNano Banana ProGoogle1K 5 · 2K 6 · 4K 101K · 2K · 4K10up to 410,000 chars
GeminiNano Banana 2Google1K 4 · 2K 6 · 4K 101K · 2K · 4K14up to 420,000 chars
GeminiNano Banana 2 LiteGoogle31K14up to 420,000 chars

These figures describe what the channels behind this platform accept and what a render costs in credits here. They are not Google's own published specification for the Gemini models behind Nano Banana.

From a prompt list to a finished grid

A family page carries one decision a single-model page does not: which Nano Banana takes the job. Make that call first, because it prices every row after it.

  1. 01

    Pick the tier before anything else

    Text-heavy layouts, charts and scenes with a cast go to Nano Banana Pro. Everyday marketing art, storyboards and diagram work run happily on Nano Banana 2. Drafts nobody will publish go to 2 Lite. One batch runs one tier, so a mixed shoot is simply two queues.

  2. 02

    Load prompts, attach references

    Paste one prompt per line, or upload a spreadsheet and point at the prompt column. All three tiers accept reference images, so a product shot or a character sheet travels with the batch and holds that likeness across rows instead of being re-described in words every time.

  3. 03

    Approve the total, then collect

    The confirmation screen shows the credit cost of the whole run before the first image starts. Approve it, watch the grid populate in place, re-queue the rows that missed, and pull the survivors out as one ZIP.

Work the Nano Banana family is genuinely good at

The family's reputation rests on two things: text that comes out readable, and subjects that stay recognisable frame to frame. Batch work leaning on either is where these models earn their credits.

Infographics and explainers

Google positions both Pro and Nano Banana 2 for diagrams, data visualisation and notes-turned-charts, and Pro's reasoning is why labels tend to land in the right cells. Queue a column of chart briefs and read the whole set side by side.

Design work carrying copy

Posters, packaging comps, greeting cards, ad variants with the headline baked in — anything where one misspelt word kills the render. Nano Banana Pro is the answer for long passages and multiple languages; Nano Banana 2 keeps most of that at a lower cost per image.

Storyboards and recurring casts

A sequence that has to keep the same faces across shots is exactly what the published likeness limits are about. Write one row per beat, attach a character reference, and get the whole sequence back in one grid.

Nano Banana questions

More questions? Email us at support@bulkimagen.com

  • No, and it is the most common mix-up about this family. Nano Banana Pro is officially Gemini 3 Pro Image, the reasoning branch from November 2025; Nano Banana 2 is Gemini 3.1 Flash Image, the speed branch from February 2026. Two positions in one generation, not two rungs of a version ladder.
  • Nano Banana Pro. Google DeepMind describes it as their best model for rendering correct, legible text, including long passages and several languages in one image. Nano Banana 2 inherits much of that at a lower cost, so drafting headlines on Nano Banana 2 and producing finals on Pro works well.
  • Google has said little beyond it being the Gemini 3.1 Flash-Lite branch, so we will not invent a capability list. What is verifiable here: it is the cheapest option on the page and covers a single output size. Use it to explore phrasings in volume, then rerun the keepers on a heavier tier.
  • Within published limits, yes. Nano Banana Pro is described as maintaining a likeness across roughly five people in a scene; Nano Banana 2 is reported at up to five characters plus fidelity across fourteen objects. Attach a reference image and keep the character wording identical in every row.
  • Yes, and the table above lists how many each will hold. Upload them once and they travel with the batch — that is the practical difference between a set of related images and a pile of unrelated ones.
  • Nano Banana Pro supports grounding through Google Search, so it can look something up before it renders, and Nano Banana 2 has access to Gemini's world knowledge. Useful for infographics with real figures. Check the numbers in a finished render anyway — grounding improves the odds, it does not guarantee accuracy.
  • Both are strong on text and structured layouts. Google gives you three price points and published consistency limits; OpenAI's model runs a reasoning pass before drawing and comes as one tier. If a shoot needs cheap drafts and expensive finals, this page is the more flexible pair of hands.
  • No. A batch runs a single model so the grid stays comparable and the credit maths stays predictable. Sending the same prompt list through twice on two tiers is a perfectly good way to compare them.
  • Nano Banana Pro is Gemini 3 Pro Image, Nano Banana 2 is Gemini 3.1 Flash Image, and Nano Banana 2 Lite is Gemini 3.1 Flash-Lite Image. The fruit names began as community shorthand before Google adopted them, which is why write-ups often disagree about what to call the same checkpoint.
  • One prompt per line in the box, or a spreadsheet upload where the prompt column becomes the rows of the batch. A catalogue export usually needs no reformatting — point at the right column and confirm what you see on the review screen.
  • New accounts are credited on signup, which is enough to put a real batch through and judge the output yourself. After that, each image costs the credits listed above for the tier and output size you chose, and the total is shown for approval before anything is charged.

Queue your first Nano Banana run

Pick the tier that fits the brief, drop in the prompt list, check the total before approving it. Nano Banana Pro for the pieces that carry text and cast, Nano Banana 2 for the volume, 2 Lite for the drafts nobody will ever see.