Nano Banana Guides Hub

Updated: 2026-08-21. Product names, available models, quotas, and access policies change. Treat Gemini, Google AI Studio, and the image generation docs as the source of truth for that day.

Introduction

Searches for “Nano Banana tutorial” or “Gemini image guide” often tangle three different things: entry (Gemini web, AI Studio, a China-friendly third-party chat), model (Nano Banana 2 / 2 Lite / Pro / legacy), and how you call it (UI clicks vs API). This page is the map for the whole Nano Banana cluster—get the layers right, then jump to model choice, China access, prompts, styles, or API without bouncing through nine articles blindly.

What this guide solves

• A single table for what to learn, which guide to open first, and what to practice next
• Clear separation of entry, model, and API so UI labels are never mistaken for API model IDs
• A mental map of the Nano Banana family: when 2, 2 Lite, Pro, or legacy actually fit
• A five-step path to a reusable first render—not one vague adjective prompt
• Scene-based routing for beginners, selection, prompts, styles, and product integration

Learning path at a glance

Goal	Read first	Practice next	Success signal
Understand what Nano Banana is	Family explainer	Same brief on 2 / Lite / Pro	You can say when Pro is unnecessary
Open and generate from China	China access guide	Bookmark a trusted entry; finish a first image	You can name third-party risks
Master the daily workhorse	Nano Banana 2 guide	Lock aspect ratio and style constraints	Fewer redo rounds
Optimize for speed and cost	Nano Banana 2 Lite	Batch drafts → refine	Clear draft vs final split
Ship your first full render	Getting started	Checklist composition and on-image text	You can reproduce the same layout
Stabilize prompts	Prompt playbook	Build a personal template library	Same prompt behaves predictably
Capture reusable styles	Styles & scenes	Style cards by use case	Team shares one description set
Wire into a product or script	API & AI Studio	Confirm model IDs in the console	Keys stay out of logs; cost is visible

Three concepts to lock in first

Entry ≠ model ≠ API

Concept	What it is	Common mistake
Entry	Web UI or workbench for login, chat, upload, and display	Thinking “opening a site” equals the latest official image model
Model	Sets quality, speed, text rendering, edit limits	Judging version from marketing copy or button nicknames alone
API	How you attach image capability to products, scripts, automation	Treating web Chat quotas as API balance

• Gemini web: gemini.google.com — preferred when you need the official product experience and public capability claims.
• Google AI Studio: aistudio.google.com — try models, tune parameters, then decide on API.
• Developer docs: ai.google.dev image generation — model IDs, rate limits, and pricing follow live docs; this guide does not hard-code numbers that age poorly.
• China-friendly convenience: Nano Banana 2 — third-party model page, not a Google product.
• Mirror / text-to-image studio: Text-to-image — multi-model switcher; review terms and routing separately.

Third-party mirrors and aggregators may differ on accounts, log retention, and model routing. Before uploading sensitive assets, trademarks, or unpublished product art, read their terms. For formal selection and billing, return to Google’s official entries and docs.

Nano Banana family map (2026)

Nano Banana is the product brand for Gemini image generation capabilities—not a separate company. Public names and API model IDs often travel in pairs; always match the docs list for that day before you write specs or code:

Product name	Common API ID (verify in docs)	One-line fit
Nano Banana 2	gemini-3.1-flash-image	Daily workhorse: speed/quality balance
Nano Banana 2 Lite	gemini-3.1-flash-lite-image	Faster and cheaper; drafts and ~1K-oriented work
Nano Banana Pro	gemini-3-pro-image	High precision, complex layouts, picky on-image text
Legacy Nano Banana	gemini-2.5-flash-image	Legacy path; new projects should move to the 2 line

Whether a model appears in your account region, renames, or reprices is always per official pages and docs that day. A “Banana” button label in a UI is not the string you must hard-code in the API.

Output quality follows constraints, not only the model tier

Under the same model, spelling out subject, lens, light, aspect ratio, and bans usually cuts rework more than blindly jumping to a pricier tier. Brand marks, lookalike faces, medical ads, and imagery of minors also need compliance and human review—smooth generation is not a green light to publish.

Treat “one generate” as a managed loop: short prompt to lock composition → add style and light → only then add on-image text. Change one variable per round so you know whether the prompt or the model tier failed. For teams, write acceptance criteria (safe margins, readable title, bans)—“make it prettier” is not actionable.

Five steps to a strong first render

1. State use and aspect: poster, cover, icon, illustration; landscape / portrait / square; target channel (web hero, short-video cover, and so on).
2. Name subject and scene: who or what sits at center; environment, time, weather, props—avoid adjective piles alone.
3. Specify style and camera: photo-real, flat illustration, isometric, cyber; framing (close / mid / wide) and light language (side light, backlight, depth of field).
4. List bans and text rules: no watermarks, no garbled English, no extra limbs; if on-image text is required, quote the exact string and ask for clear, readable glyphs.
5. Human-check, then iterate: match use case for composition, readability, brand consistency; change one variable (light or framing) so failures are diagnosable.

Copy-ready first-image template:

Use: [web hero / cover / product illustration]
Aspect: [16:9 / 1:1 / 9:16]
Subject: [clear person or object, including material and color]
Scene: [place, time, mood]
Style: [photo-real / flat illustration / other concrete style]
Camera: [framing + lighting]
On-image text (if any): "exact copy"; must be sharp and typo-free
Ban: watermarks, extra fingers, warped logos, cluttered background text
Output: one primary image plan; if anything is unclear, ask before guessing brand rules.

Second prompt (quick entry comparison):

Generate a photo-real still life: a coffee cup and a closed notebook on a desk.
Natural window light, shallow depth of field, no watermark, no close-up faces.
Do not add any on-image text.

Run the same prompt on official Gemini, AI Studio, and any China-friendly third-party entry you have evaluated. Log speed, quality, and whether the model invents text—then pick a long-term path.

Topic map by scenario

What you want now	Open this	Have ready
First concepts	What is Nano Banana	Three real image briefs
Official site won’t open / need a fast try	China access	A privacy-safe test prompt (no trademarks)
Systematically master 2	Nano Banana 2	Fixed aspect + style card
Cheap batch drafts	2 Lite	Draft → refine two-stage flow
Hand-held first image	Getting started	This page’s five-step template
Prompts keep collapsing	Prompt playbook	Personal negative list
Reusable style library	Styles & scenes	Brand colors + licensed refs
Product integration	API & AI Studio	Current model names from official docs

Most of the time, fix the prompt and acceptance bar before upgrading to Pro. Blind model swaps make bugs harder to reproduce.

Access & entries

• Official Gemini: gemini.google.com
• Google AI Studio: aistudio.google.com
• Developer docs: ai.google.dev · image generation
• China-friendly third-party · Nano Banana 2: Nano Banana 2
• Mirror / text-to-image studio: Text-to-image

Start with a fixed, privacy-safe test prompt (no unauthorized trademarks). Compare speed, composition obedience, and text rendering. Before uploading enterprise assets, portraits, or unpublished designs, confirm terms and your organization’s data policy.

Startup checklist

• You can separate entry, model, and API—and never treat a page title as an official API ID
• You bookmarked at least one trusted entry (official Gemini / AI Studio, or a third-party entry whose terms you assessed)
• You completed one “five-step render” with a privacy-safe fixed prompt and human review
• You know whether today’s job is selection, prompts, style capture, or product wiring—and opened the matching guide
• You treat every key frame as needing human verification; first generate is never the final asset by default
• Before formal selection or purchase, you cross-check Google’s official pages and docs—not only third-party button labels

FAQ

Is Nano Banana free?

Free tiers, queues, and paid plans for Gemini web, AI Studio, and the API all change. Follow the live product and billing pages—not outdated screenshots or second-hand quotes.

Nano Banana 2 vs Pro—how do I choose?

For daily heroes, social art, and most marketing drafts, start with Nano Banana 2. Evaluate Pro when you need extreme detail, complex typography, or print-oriented output. When cost and latency matter and you are exploring layouts in bulk, draft on 2 Lite first. Details: family explainer.

Is a China third-party entry the same as official?

No. Third parties own reachability and packaging; accounts, logs, routing, and support channels may differ. For capability claims and API parameters, return to Gemini and the developer docs.

Can I use generated images commercially?

It depends on the product terms you use, regional rules, asset copyright, and what appears in the frame (trademarks, likeness, brand IP). Treat the model as a draft accelerator; finish rights and compliance review before commercial use. This guide is not legal advice.

Where should I start reading?

Complete beginners: this hub → What is Nano Banana → China access or Getting started. Stable entry already, quality only: go straight to the prompt playbook.

Official resources

• Gemini
• Google AI Studio
• Google AI for Developers
• Gemini image generation docs

Further reading

• What is Nano Banana? Family explainer
• Nano Banana in China: full access guide
• Nano Banana 2 complete guide
• Getting started: first render
• Prompt playbook
• API & AI Studio intro

Summary

The right order is: trusted entry → pick 2 / Lite / Pro by task → control quality with structured prompts and human review. Use this page as a table of contents: finish one real image job, then open the matching deep dive—faster than reading every article in one sitting. Entry is not model; third-party is not official; prices and model IDs that age poorly always follow Google’s docs for that day.