A good Gemini image prompt names the subject, what it is doing, where it is, the visual style, the lighting and the camera or format, written as a natural description rather than a list of keywords. This gemini prompt cheat sheet gives you that formula, ready-to-edit templates for common image types, a set of editing commands and the mistakes that most often produce poor results.
The example prompts below are my own templates, not tested outputs. Results vary by model, session and wording, so treat them as starting points.
Which Gemini model creates your images?
Gemini generates images with Google's "Nano Banana" family of models, so prompting advice for "Nano Banana" applies directly to Gemini. Nano Banana is a nickname used online for Gemini's image-generation model. The family has grown since launch. According to one third-party guide, as of mid-2026 there are three versions:
- Nano Banana, built on Gemini 2.5 Flash Image. Google announced it on 26 August 2025. It suits quick drafts.
- Nano Banana 2, built on Gemini 3.1 Flash Image. That guide calls it the default choice for most everyday image tasks.
- Nano Banana Pro, built on Gemini 3 Pro Image. Google's developer documentation, as summarised in a technical write-up, describes it as designed for professional assets, using advanced reasoning to follow complex instructions and render high-fidelity text.
Model names, labels and availability change often. Check the model shown inside your Gemini app before relying on any comparison, including this one.
how to use AI to supercharge your resume
What is the basic Gemini image prompt formula?
Use this six-part formula: subject + action + setting + style + lighting + camera or format. You do not need all six every time. Add parts until the image matches your intent.
| Element | What to specify | Example wording |
|---|---|---|
| Subject | Who or what, with defining details | "an elderly tea seller with a grey moustache" |
| Action | What is happening | "pouring chai into a clay cup" |
| Setting | Place, time, background | "on a rainy Mumbai street at dusk" |
| Style | Photo, illustration, 3D, painting | "documentary photograph" or "flat vector illustration" |
| Lighting and mood | Light source and feeling | "warm streetlight, soft reflections, calm mood" |
| Camera or format | Shot type, lens, aspect ratio | "85mm portrait lens, shallow depth of field, 4:5" |
One third-party guide that cites Google's official prompting guidance describes a similar pattern: a request to create an image, then subject, action, scene and style details.
Describe the scene in sentences. An independent review of Google's official guidance says it comes down to two text templates and one narrative principle. It also notes that a widely shared "10-command framework" is not an official Google guide. Be wary of any cheat sheet that claims Google's authority without linking to the source.
Gemini prompt templates you can copy
Replace the bracketed parts and keep the sentence structure. Each template is built for a specific image type.
Photorealistic scene
A photorealistic [shot type] of [subject], [action or expression], set in [environment].
Lit by [light source], creating a [mood] atmosphere.
Shot with [lens or camera detail], emphasising [textures or details].Example: A photorealistic close-up of a ceramic mug of masala chai on a wooden windowsill, steam rising, set in a monsoon-lit apartment. Lit by soft overcast daylight, creating a quiet, cosy atmosphere. Shot with a 50mm lens, emphasising the glossy glaze and condensation on the glass.
Example: A photorealistic close-up of a ceramic mug of masala chai on a wooden windowsill, steam rising, set in a monsoon-lit apartment. Lit by soft overcast daylight, creating a quiet, cosy atmosphere. Shot with a 50mm lens, emphasising the glossy glaze and condensation on the glass.
Illustration or sticker
A [style] illustration of [subject] with [key features].
Use [colour palette] and [line style].
Background: [plain / white / transparent-style].
Example: A flat vector illustration of a friendly robot watering a balcony garden, with rounded shapes and big eyes. Use a pastel green and peach palette with thick outlines. Background: plain white.
Example: A flat vector illustration of a friendly robot watering a balcony garden, with rounded shapes and big eyes. Use a pastel green and peach palette with thick outlines. Background: plain white.
Product mockup
A high-resolution product photograph of [product] on [surface],
[lighting setup], [camera angle], clean background,
space on the [left/right] for text.Text inside an image
A [poster / logo / label] with the exact text "[YOUR TEXT]" in [font style],
[colour], placed [position]. Design style: [style].Tip: put the exact wording in quotation marks and keep it short. Always proofread the output, because long or unusual strings can still be misspelt.
Minimalist composition
A minimalist composition featuring [subject] positioned in the [lower-right] of the frame,
surrounded by large areas of empty [colour] space. Soft, even lighting.Comic panel or storyboard
A [number]-panel comic strip in [style].
Panel 1: [scene]. Panel 2: [scene]. Panel 3: [scene].
Consistent characters: [short description of each].
prompt cheat sheet for social media posts
How do you edit an image in Gemini?
Upload the image, then describe the single change you want, and keep everything else unchanged by saying so. Editing in the same chat lets you refine step by step. Google's own example prompts include asking for a follow-up scene with the same character, adding objects to an existing picture, and converting an image to a blueprint style. These include "Convert this image to an architectural blueprint style" and "add three stacked books on the coffee table."
Editing command cheat sheet
| Goal | Prompt pattern |
|---|---|
| Add an object | "Add [object] to [location]. Keep everything else the same." |
| Remove an object | "Remove the [object]. Fill the gap naturally." |
| Change style | "Convert this image to [style]. Keep the composition." |
| Change background | "Replace the background with [new setting]. Keep the subject unchanged." |
| Change lighting or time | "Make this scene look like [golden hour / night]." |
| Combine images | "Place the person from image 1 into the scene from image 2." |
| Keep a character consistent | "Show the same character in [new scene]." |
Edit one thing at a time. Stacking six changes in one request raises the chance that something unrelated shifts. If an edit goes wrong, say what went wrong: "The face changed. Restore the original face and only change the jacket colour."
why are people leaving chatgpt
Which style and lighting words work well?
Specific, concrete words give better control than vague praise such as "beautiful" or "stunning." Use this quick-reference list and mix as needed.
| Category | Useful terms |
|---|---|
| Photo style | documentary, editorial, macro, street photography, aerial |
| Illustration | flat vector, watercolour, line art, pixel art, retro poster |
| Lighting | golden hour, overcast, rim light, studio softbox, neon glow |
| Camera | wide-angle, 85mm portrait, shallow depth of field, low angle |
| Mood | calm, nostalgic, dramatic, playful, minimal |
| Composition | centred, rule of thirds, negative space, symmetrical |
Several of these terms (line art, pixel art, oil painting, retro poster style) appear in public Gemini prompting guides as style options for Nano Banana prompts.
Common Gemini prompt mistakes
- Keyword piles. "Dog, park, 4k, ultra, masterpiece" gives the model nothing to interpret. Write a sentence.
- Vague adjectives. Replace "nice lighting" with "soft window light from the left."
- Too many ideas in one prompt. Pick one subject and one setting.
- Asking for long text. Short phrases are safer than paragraphs.
- Not iterating. The first image is a draft. Refine in the same chat.
- Ignoring aspect ratio. State it up front (square, 16:9, 9:16) if you need a specific format.
- Copying viral prompts blindly. Adapt them to your subject instead of pasting.
What are the limits to know about?
Gemini images have usage limits, regional differences and a built-in watermark. Three points matter most:
- SynthID watermark: Documentation summarised in a technical write-up says all generated images include a SynthID watermark.
- Usage limits and features vary by plan, country and date. Check Google's current help pages instead of trusting a blog's figures.
- Content policies apply. Requests for harmful, deceptive or rights-infringing images may be refused. Avoid using real people's likenesses without permission.
Who is this cheat sheet for?
It suits beginners who want better first results and marketers, designers and creators who need repeatable prompt patterns. It is less useful if you need exact brand colours or pixel-perfect layouts. For those, generate a base image in Gemini and finish it in a design tool.
Key Takeaways
- Gemini image generation runs on the Nano Banana model family. Check which version your app uses.
- Use the six-part formula: subject, action, setting, style, lighting, camera or format.
- Write full descriptive sentences, not keyword lists.
- Copy the templates above and replace the bracketed parts.
- Edit one change at a time, and tell Gemini what to keep unchanged.
- Put exact text in quotation marks, keep it short and proofread it.
- Iterate in the same chat instead of starting over.
- Usage limits, features and model names change, so verify them on Google's official pages.
Frequently Asked Questions
What is a Gemini prompt cheat sheet?
It is a quick reference of prompt formulas, templates and editing commands for generating images in Gemini. A good one gives you a repeatable structure, such as subject, action, setting, style, lighting and camera, instead of a list of random keywords.
Is Nano Banana the same as Gemini image generation?
Yes. Nano Banana is the popular nickname for Gemini's native image model family. Names such as Nano Banana 2 and Nano Banana Pro refer to different versions with different speed and quality profiles.
How long should a Gemini image prompt be?
Long enough to describe the subject, setting, style and lighting, usually two to four sentences. Very long prompts with many competing ideas tend to dilute the result. Start simple and add detail in follow-up edits.
Can Gemini put exact text in an image?
It can, and newer models handle text better than older tools, but errors still happen. Keep the text short, put it in quotation marks, name the font style and proofread the output.
How do I keep the same character across images?
Stay in the same chat, describe the character briefly each time and ask for "the same character" in a new scene. You can also upload a reference image. Results are not guaranteed to be identical.
Are Gemini-generated images watermarked?
According to Google's developer documentation as summarised by a technical source, generated images carry a SynthID watermark. Check Google's current policy before commercial use.
Do I need to pay to use Gemini image generation?
Free access has been reported, but limits and features change by plan and region. Check the Gemini app or Google's help pages for current details.
Post a Comment