What is a Nano Banana prompt?
Nano Banana is Google's nickname for Gemini's image models. A Nano Banana prompt is the text you paste into the Gemini app or Google AI Studio. Google's prompting guide for a new image is a plain sequence: subject, action, location, composition, and style. If words have to appear in the picture, they go in quotes. You can name an aspect ratio in the sentence. Resolution, such as 1K or 2K, is a control in the product, not a phrase the sentence has to smuggle in.
Reprompt (reprompt.org) writes that prompt for you. Give it a short idea, a reference photo, or both. You get a sentence and a JSON copy of the same decisions. Free, no account, with a pace limit and a daily ceiling. This page does not generate the image. Copy the prompt into Gemini or AI Studio and run it there.
When do you write a new prompt, and when do you write an edit?
Create mode is for a picture that does not exist yet. The sentence should be able to stand alone. Name the object, what it is doing, where it is, how the frame is arranged, and the style. "A ceramic mug" is a subject. "A ceramic mug on a sunlit wooden windowsill, centered, product photo, soft side light" is a prompt.
Edit mode is for a photo you already have. The prompt is an instruction: what to change, then what must stay. Google's guide treats the uploaded photo as the reference and the text as the change. If you attach the photo here, the instruction can name what is actually in the frame. You still upload that photo in Gemini or AI Studio yourself. Reprompt only writes the words.
If you want the photo described as fields first, use image to JSON prompt. If you want a video move from the same still, use image to video prompt. The Nano Banana guide is the longer background on the model.
Should a Gemini image prompt be text or JSON?
Start with the text prompt. Gemini's image box reads language. Google does not publish a special JSON prompt language for Nano Banana, and the consumer prompt box does not have a JSON switch. The JSON on this page is an organizing copy: subject, action, location, composition, style, lighting, camera, any text to render, and the aspect ratio. Edit mode also fills "edit" and "preserve."
Use the JSON when you want to change one decision, such as the light, and then paste either the object or a sentence you rewrote from it. If a run ignores a field, that is a known way these prompts fail. Switch to the sentence and put that detail in words. Do not keep adding braces.
| Create | Edit | |
|---|---|---|
| You have | An idea, and maybe a reference for style. | A photo, and a change you want. |
| The prompt says | Subject, action, location, composition, style. | What to change, then what must stay the same. |
| Where it runs | Gemini or AI Studio, as a new image. | Gemini or AI Studio, with your photo uploaded there. |
| What this site does | Writes the prompt. | Writes the instruction. It does not edit the file. |
What should the prompt refuse to do?
Reprompt doesn't produce prompts for sexual or explicit images, minors, or non-consensual edits of real people. An edit that changes the background, the crop, or the color of a product is a normal request. An edit that undresses someone, fakes an intimate photo, or puts one person's face on someone else is refused. If you are describing your own photo, say what must stay, including the face, so the instruction does not wander.
Quoted text is for words you want drawn, such as a short label on a mug. Keep it short. The model is better at a few words than at a paragraph of type. After you have the sentence, the prompt rewriter can aim the same idea at Midjourney or Stable Diffusion if you want a second tool.
What does a real Nano Banana prompt look like?
This is one create-mode run from October 6, 2026. No photo was attached. The idea was "a ceramic mug of coffee on a sunlit wooden windowsill, product photo, no text". The sentence follows the order in Google's guide closely enough to paste: subject, what it is doing, where, how it is framed, and the style. The aspect ratio landed in the sentence as 4:5. textToRender is empty because the idea said "no text." Edit and preserve are empty because this was not an edit.
Text prompt from that run
A ceramic mug of freshly brewed coffee resting on a rustic wooden windowsill, bathed in warm, soft sunlight, captured in a clean product photography style with a shallow depth of field in a 4:5 aspect ratio.
JSON from that run
{
"subject": "ceramic mug of coffee",
"action": "resting on a windowsill",
"location": "wooden windowsill",
"composition": "product photography, shallow depth of field",
"style": "clean product photography",
"lighting": "warm, soft sunlight",
"camera": "macro lens",
"textToRender": "",
"aspectRatio": "4:5"
}Paste the sentence into the Gemini app or Google AI Studio. Paste the JSON only if you want to change one field first, then paste that text into the same box. Neither version generates an image on this site. If you switch to edit mode with a photo of your own mug, the prompt should name the change and what to keep, and you upload the photo in Gemini, not here.
Nano Banana prompt questions
What is a Nano Banana prompt?
A Nano Banana prompt is the text you paste into Gemini's image model, which Google nicknames Nano Banana. Google's prompting guide builds a new image from a subject, an action, a location, a composition, and a style. Reprompt (reprompt.org) writes that prompt from your idea, from a reference photo, or from both.
Does this page generate the image?
No. Reprompt writes the prompt. You paste it into the Gemini app or Google AI Studio, where Nano Banana runs. Nothing on this page calls an image generator.
Should I paste the text prompt or the JSON?
Paste the text prompt when you want Gemini to read a sentence. Paste the JSON when you want to edit one field first. Google does not publish a separate JSON mode for Nano Banana. Both versions are text in the prompt box. If a field gets ignored, switch to the sentence.
How do I write an edit prompt for a photo I already have?
Choose edit, describe the change, and upload the photo if you want the instruction to name what is in the frame. The result says what to change and what must stay. You still upload that photo yourself in Gemini or AI Studio. This page only writes the instruction.
What is a Gemini image prompt?
It is the same kind of prompt. Nano Banana is the public nickname for Gemini's image models, including Nano Banana 2 (Gemini 3.1 Flash Image) and Nano Banana Pro. A Gemini image prompt names the subject and the change in ordinary language. Put any words that must appear in the picture in quotes. Set resolution in the product. The prompt can mention an aspect ratio such as 1:1 or 4:5.
Will you write a prompt that edits a real person without consent?
No. Reprompt doesn't produce prompts for sexual or explicit images, minors, or non-consensual edits of real people. That includes undressing someone, a fake intimate photo, and swapping one person's face onto someone else.
Related tools: image to JSON prompt, image to prompt, prompt rewriter, and image to video prompt, the JSON prompt generator, and the art style identifier.