How to prompt Nano Banana for best results?
Describe a full scene in plain sentences, not keyword lists. Name the subject, setting, lighting, camera angle, and mood, then refine conversationally. Google's own guidance is that the more specific detail you add, the closer the result. Upload reference images to lock character or product likeness, and edit by asking for one change at a time.
Why — the first-principles explanation
Nano Banana is built on Gemini, a language model, so it understands full sentences and real-world context far better than the comma-separated keyword tags older image tools relied on. When you write 'a golden retriever puppy sitting on a red picnic blanket in soft morning light,' the model reasons about each relationship — the dog is on the blanket, the light is morning — the way a person reading the sentence would. Vague prompts leave those decisions to chance.
Because it is conversational, the best workflow is iterative. You generate a first image, then talk to it: 'make the light warmer,' 'turn the puppy to face left,' 'add a picnic basket.' The model keeps the rest of the scene stable and applies just your change, so you sculpt the picture step by step instead of rewriting one giant prompt.
Nano Banana also accepts images as input, not just text. Uploading a reference photo of a person, product, or style tells the model exactly what to keep consistent — this is how it holds a character's face across many pictures. Google's prompt guidance stresses describing intent and providing detail; the Pro model even reasons about legible text, so you can literally ask for a poster with specific words on it.
The core principle: specificity plus dialogue. Say what you want fully, show a reference if you have one, then adjust in small conversational steps rather than starting over.
An example that makes it click
Imagine ordering a custom cake. If you say 'a cake,' you get whatever the baker guesses. If you say 'a two-layer chocolate cake with white frosting, seven candles, and blue roses on top, photographed from above,' you get your cake. Then instead of throwing it out to change one thing, you just tell the baker 'make the roses pink' — and everything else stays the same. That's exactly how you talk to Nano Banana: describe fully, then tweak one detail at a time.
How to do it
- Write a complete sentence naming the subject, setting, lighting, camera angle, and mood.
- Add specific details — colors, materials, time of day, art style — since more detail means closer results.
- Upload a reference photo if you need a consistent face, product, or style.
- Generate, then refine with one conversational instruction at a time ('make it warmer,' 'remove the hat').
- For posters or mockups, use Nano Banana Pro and specify the exact text you want rendered in the image.
- Save prompts that work so you can reuse and lightly vary them.
Key facts
- Nano Banana is built on Gemini and interprets natural-language sentences, not keyword tags.
- Google's guidance: the more detail you add, the closer the image is to what you imagined.
- The model supports conversational, iterative editing that changes one element while preserving the rest.
- Reference images can be uploaded to lock character, product, or style consistency.
- Nano Banana Pro can render legible, multilingual text directly in images for posters and mockups.
Google's Gemini image model (nicknamed Nano Banana), known for consistent edits.
Affiliate link — we may earn a commission at no cost to you.▶ The 60-second explainer (script)
Want better images from Nano Banana? Stop typing keyword lists and start writing sentences. Because Nano Banana runs on Google's Gemini, it understands full descriptions the way a person does. So instead of 'dog, blanket, sunlight,' say 'a golden retriever puppy on a red picnic blanket in soft morning light, photographed from above.' The more specific you are, the closer you get. Next trick: it's conversational. Don't rewrite your whole prompt to fix one thing — just tell it 'make the light warmer' or 'turn the puppy to the left,' and it keeps everything else the same. Got a specific face or product you need to match? Upload a reference photo and it'll stay consistent. And if you're using Nano Banana Pro, you can even ask for a poster with exact words spelled correctly. Describe fully, show a reference, then tweak one step at a time.
What authoritative sources say
People also ask
Should I use commas and keywords like in Midjourney?
No. Nano Banana works best with full natural-language sentences because it is built on the Gemini language model.
How do I keep the same character across images?
Upload a reference photo of the person and ask for new poses or scenes; the model preserves their likeness.
Why is my text coming out garbled?
Use Nano Banana Pro for text, and state the exact words in quotes; the Pro model is tuned for legible, multilingual text.
What's the fastest way to fix a small flaw?
Give one conversational edit like 'remove the sunglasses,' which changes just that detail while keeping the rest intact.