Nano Banana
Nano Banana 2 Lite Prompt Guide: Tips and Examples
Nano Banana AI4 min read

Contents+
If you want steadier first frames from Nano Banana without writing novels, this is the short path. The Nano Banana 2 Lite Prompt Guide keeps the habits that matter for Nano Banana 2 Lite text-to-image work—structure over adjective stacks, one goal per line, then chat polish.
What a Lite prompt approach means
Lite here is not a secret second model. It is a lighter Nano Banana / Nano Banana 2 writing habit:
- Prefer structure over filler adjectives
- Aim for a directionally correct first frame, then conversational edits
- Let each prompt serve one clear job
Built on Gemini 2.5 Flash Image-style generate-and-chat, it fits daily concepts, covers, product stills, and fast scouting.
Try it on nanobanana-ai.us/en/ in the Nano Banana workbench.
The five-part formula
For most text-to-image prompts, write in this order:
- Subject
- Scene
- Style (one primary lane)
- Lighting
- Lens / framing
Remember: clear subject → grounded light → single style.
Blank template
[subject], [scene], [style], [lighting], [lens feel]
Five prompt tips you can use today
1. Name what it is before how pretty it looks
Weak: beautiful amazing masterpiece coffee
Strong: ceramic coffee pot on a walnut table, tall-window side light, shallow depth of field, photoreal product photography
2. Keep one primary style per prompt
“Photoreal + anime + oil paint” fights itself. Lock one lane; save hybrids for a later chat turn.
3. Lighting beats empty filter words
Prefer side light, overhead, soft, hard, dawn, dusk, neon spill. Skip vague cinematic 8k unless you also name direction and quality.
4. First frame only needs the right direction
Subject, crop, and light mood can be “good enough.” Detail converges with Nano Banana conversational edits—not a paragraph-long first prompt.
5. Change one thing per edit turn
Keep the subject unchanged. Replace only the background with a dusk shoreline.
One focus per turn matches how Nano Banana 2 stays stable across chat.
Example library (rewrite freely)
A. Product still (commerce)
Glass skincare bottle on pale marble, pump and label readable, soft window light, shallow depth of field, ecommerce hero, photoreal product photography
B. Portrait mood
Short-haired woman on a rain-washed city street, 85mm portrait, neon reflections, wet asphalt, natural expression, filmic contrast
C. Interior
Modern living room corner with armchair and floor lamp by the window, afternoon side light, clean negative space, architectural interior photo, mild wide angle
D. Cover / editorial still
Open notebook and pen on a desk, morning window light, warm grade, minimal still life, magazine cover framing
E. Chat edit after the scout
Keep product shape and label unchanged. Remove extra props on the right. Soften the background blur a little more.
Pitfalls vs Lite fixes
| Pitfall | Better Lite move |
|---|---|
| Adjective pile, no subject | State who/what and where first |
| Mixed style lanes | Keep one primary style |
| Long unstructured prompt | Compress back to the five parts |
| Wrong direction, ten chat fixes | Redraw if the scout is wrong; chat only when direction is right |
| Five changes in one turn | One change per turn |
Ten-minute drill
- Write one product prompt with the formula; generate.
- Write one portrait or interior prompt; generate.
- On a passable scout, run one single-point edit.
- Save the winning sentence as a labeled template.
Workbench:
https://app.nanobanana-ai.us/en/generate/image-tools/nano/
How this links to the handbook
- Want the full map from first prompt to publishable frame? Read The Nano Banana 2 Handbook.
- Want deeper prompt craft? Pair this Lite guide with the site’s prompt tips.
- Want when to redraw vs chat-edit? See Nano Banana vs traditional AI art.
Lite is the daily muscle; the handbook is the route map.
Wrap-up
Lite prompting is not a longer spell. It is a lighter structure for Nano Banana / Nano Banana 2:
- Five-part formula for the scout
- One style + named lighting
- One change per conversational turn
Use that rhythm on text-to-image jobs and you reach “editable and publishable” sooner. Open nanobanana-ai.us/en/, grab any example above, rewrite it once, and generate.


