Nano Banana 2 is currently positioned as a strong all-round free-tier recommendation, reached through the Gemini app. What makes it a good first tool is not any single capability but the combination: it is fast, it produces believable images, and it lets you correct your way to a result in ordinary sentences.
This tutorial assumes nothing. By the end you will have generated a portrait, a product shot, a travel scene and a social graphic, and you will know how to edit an image without losing the parts you liked.
Key takeaways
- Write the sentence you would say to a competent assistant. Formal prompt syntax is unnecessary here.
- Iterate in the same conversation. The context carries, so you can change one thing at a time.
- Constraint clauses matter: "keep everything else identical" prevents unwanted drift.
- It is strong on realism and speed, weaker on strongly art-directed stylisation and precise typography.
- Free-tier limits, regional availability and features can change without notice.
Getting started in Gemini
Open the Gemini app, sign in, and start a new conversation. Ask for an image in the same way you would ask for anything else — there is no separate mode to learn and no parameter syntax to memorise.
Before you generate anything, decide what the image is for. A one-line brief such as "a warm header image for a newsletter about slow mornings" gives you a standard to judge results against. Without it, you will accept the first pleasant image you see.
Check what your free allowance covers before starting a longer session. Allowances and availability vary and can change, and it is frustrating to run out halfway through a refinement sequence.
Writing your first prompt
Cover four things: subject, scene, composition and light. That structure comes from our pillar guide to AI image generation and works everywhere, including here.
Keep the language natural. "A woman reading by a window on a rainy afternoon, shot from across the room on a 50mm lens" is better than a list of comma-separated keywords, because this model handles sentences comfortably.
A portrait of a man in his sixties with a short grey beard, sitting at a kitchen table with a mug of tea. Soft daylight from a window to his left, natural skin texture with visible lines, 50mm lens, waist-up framing, calm and unposed, muted natural colour.
"Natural skin texture" and "unposed" do most of the work here. Both push against the default polish.
Iterative prompting and editing
The real advantage of a conversational tool is the second message. Rather than rewriting your prompt, describe the single change you want and state what must stay the same.
Good edit instructions are specific and bounded. "Move the camera lower and slightly to the right, keep the lighting and the subject exactly as they are" will usually hold. "Make it better" will not.
Drift is the main hazard. Each edit is a fresh generation informed by context, so small differences accumulate across five or six rounds. Restate your non-negotiables each time, and start a new conversation when a series has wandered too far.
Using this image, change only the background to a plain warm grey studio wall. Keep the subject, pose, clothing, lighting direction and colour grade exactly as they are. Do not change the framing.
Name what changes, then name what must not. The second clause is the one people forget.
Five practical recipes
These cover the work most people actually need. Adapt the nouns; keep the structure.
A matte black water bottle standing on a pale concrete surface, photographed for a product page. Single soft light source from the upper left, gentle shadow to the right, no reflections on the label, plain background, 85mm lens at f/8, honest colour, nothing stylised.
"Nothing stylised" is a useful brake when you need a catalogue image rather than an advertisement.
A narrow cobbled street in a hillside town in southern Italy, early morning, shutters still closed, one cat on a doorstep, laundry overhead. Soft directional sunlight down the length of the street, 35mm lens, slightly imperfect framing, natural colour with no HDR effect.
"No HDR effect" removes the over-processed look that travel prompts often attract.
A square image for social media: a flat overhead arrangement of gardening tools, seed packets and soil on a weathered wooden bench, plenty of empty space in the upper third for a headline to be added later, soft even daylight, muted earthy palette, no text in the image.
Generate the background clean, then typeset the headline in a design tool. Far more reliable than asking for text.
| Output | Emphasise | Avoid |
|---|---|---|
| Portrait | Skin texture, single light source, unposed | "Perfect", "flawless", heavy retouch language |
| Product | Material behaviour, controlled shadow, plain ground | Dramatic lighting, busy props |
| Travel | Time of day, weather, one human detail | Saturation boosts, HDR |
| Social | Negative space, strong focal point | Requesting rendered text |
| Edit | One change plus explicit constraints | Multiple simultaneous changes |
Common mistakes
The first is asking for quality instead of content. Words like "masterpiece" and "8k" describe your hopes, not the picture. Replace them with a lens, a light and a material.
The second is changing five things at once, which makes it impossible to know what helped. The third is expecting rendered text to be correct; keep copy out of the generation and add it afterwards.
The fourth is treating a free tier as production infrastructure. If a deadline depends on a specific model being available in your region on a specific day, build in a fallback.
Availability can change
Free-tier limits, model versions and regional availability for the Gemini app can change at any time. Verify current terms before relying on this workflow for client or commercial work.
When it is not the right tool
For a distinctive art-directed look, Midjourney is usually the better start. For posters and thumbnails where words must be exact, Ideogram 3 leads. For technical photorealism and API pipelines, FLUX.2 is often preferred.
For game assets, character sheets and sketch-to-image work, Leonardo AI is purpose-built, and our Leonardo AI tutorial covers those workflows in detail.
None of that diminishes Nano Banana 2. It is the tool we would put in a beginner's hands first, and many people never need anything else.
Frequently asked questions
Is Nano Banana 2 free to use?
It is available through the Gemini app with a free tier. Allowances, regional availability and feature sets can change, so treat free access as a convenience rather than something to build a business process around.
What is Nano Banana 2 best at?
It is often preferred for speed, believable realism, ease of use and conversational editing. Everyday photography-style images, simple product scenes, portraits and social content are its comfortable range.
Can I edit an image I already have?
Yes. Upload the image and describe the change in plain language. Constraint-led instructions such as "change only the background, keep the lighting identical" work considerably better than open-ended requests.
Why do my results drift when I keep editing?
Each edit is a new generation informed by the conversation, so small differences accumulate. Restate the parts that must not change in every message, and start a fresh conversation when drift becomes severe.
When should I use something other than Nano Banana 2?
Choose Midjourney for strong art direction, Ideogram 3 for readable text, FLUX.2 for technical photorealism and API work, and Leonardo AI for game assets and consistent character sets.
Return to the pillar guide for the full picture: how these systems work, how the model landscape fits together, and which tool suits which job.
Continue the series
The other chapters in the AI Image Generation guide.