How to Write Nano Banana Prompts: A Stable First Frame in 15 Minutes
Beginner guide to Nano Banana prompts: use a six-part formula to write clearer AI art prompts and land a stable first frame in about 15 minutes.
This tutorial focuses on one skill: writing Nano Banana prompts so a beginner can get a directionally correct first frame in about 15 minutes. Ideal if you are new to Nano Banana (Gemini 2.5 Flash Image) for text-to-image / AI art.
Learning Goals
By the end you will be able to:
- Write one clear prompt with a fixed structure
- Decide whether a first frame is “good enough” instead of regenerating blindly
- Apply one short edit instruction for a minimal fix
What You Need
- Access to the Nano Banana workbench (start from nanobanana-ai.us)
- A simple subject (coffee cup, product bottle, window portrait—anything concrete)
- About 15 minutes
Workbench entry:
https://app.nanobanana-ai.us/en/generate/image-tools/nano/
Step 1: Memorize the Prompt Formula (2 Minutes)
When writing AI art prompts, do not start by stacking adjectives. Write in this order:
- Subject (the person or object that matters most)
- Action / state (optional)
- Scene / environment
- Style (photoreal, illustration, product photography, etc.)
- Lighting (direction, soft/hard, time of day)
- Lens / framing (close-up, wide, eye-level, shallow depth of field)
Mantra: clear subject → grounded light → one style. That beats “beautiful, 8k, masterpiece” for Nano Banana.
Step 2: Fill the Template (3 Minutes)
Copy this skeleton and replace the brackets:
[subject], [scene], [style], [lighting], [camera feel]
Example A: Product still life
Ceramic pour-over coffee pot on a walnut table, high-window side light, soft shadows, shallow depth of field, realistic product photography
Example B: Portrait mood
A short-haired woman standing on a rain-washed street, 85mm portrait, neon reflections, wet asphalt, natural expression, cinematic contrast
Before you generate, check: would a stranger instantly know who/what the subject is, where they are, and what the light feels like? If not, fix subject and scene first—do not pile on style words.
Step 3: Generate and Run the Three-Question Check (5 Minutes)
Paste the prompt in the workbench and generate 1–2 images. For each frame, ask only:
- Is the subject correct?
- Is the composition usable?
- Does the lighting mood feel close?
If two of three pass, keep it. If all fail, regenerate. Do not grind ten edits on a wrong direction—that is the most common time sink when practicing Nano Banana prompts.
Step 4: One Short Edit (4 Minutes)
Once the first frame points the right way, change one thing only, for example:
Keep the subject unchanged. Change only the background to a dusk shoreline.
or:
Soften the key light a bit; leave everything else as unchanged as possible.
Good text-to-image prompts set direction; short conversational edits tighten the result.
Step 5: Save Your Personal Template (1 Minute)
Store today’s working prompt as a template and label the use case (product / portrait / cover). Next time, swap only subject and scene—reuse style and lighting. That habit is how Nano Banana prompts get more stable over time.
Common Mistakes
| Weak approach | Stronger approach |
|---|---|
| Vague adjective piles | Name subject, materials, light direction |
| Mixing three styles at once | Pick one primary style |
| Long unstructured walls of text | Compress with the six-part formula |
| Changing five things per turn | Change one priority per turn |
Practice Checklist (Finish Today)
- Write 1 product-style prompt with the formula and generate
- Write 1 portrait/scene prompt with the formula and generate
- Apply 1 single-point conversational edit to a passing first frame
- Save 1 reusable template
Wrap-Up
How do you write Nano Banana prompts? With structure, not superstition: subject → scene → style → lighting → lens. Your beginner goal is not one-shot perfection—it is a stable first frame in 15 minutes, then conversational polish until it is publishable.
Next, try the on-site conversational editing and character consistency tutorials—or return to the workbench and run two more subjects with the same formula.