How to use Nano Banana prompts on your own photos

Nano Banana is the nickname of the image model inside Gemini behind a huge share of prompt seen edits: the ones with bold colour, festival lighting and a face that stays recognisably yours. Using it is not about prompt-writing wizardry. It is about naming six things in one paragraph: the person from your reference photo, the scene, the outfit, the pose, the lighting and the quality. This post walks through the whole flow, from photo choice to download.

Step 1: start with a photo the model can read

The render keeps the face it is given, so the upload sets the ceiling. One clear, front-lit face photo, no sunglasses, no heavy filters. Slight angles are fine; deep side profiles are not. If you are doing a couple edit, prepare one photo per person rather than cropping a group picture.

Step 2: write or borrow the prompt

You can write your own or copy one from the prompt shelf. The structure is always the same, and a Diwali example shows the shape:

Place the man from the reference photo in a courtyard decorated with rangoli and marigold torans. He wears a black silk kurta with gold buttons, hands folded in namaste, calm half smile. Golden hour light mixed with diya flames, gentle rim light on his hair, festive poster finish, ultra detailed, 4:5.

Notice the order: who, where, wearing what, doing what, under which light, at which quality. Swap any noun and you have a new edit without touching the structure.

Step 3: run it in the studio

Open the studio, pick the look or paste your prompt into the prompt box, upload the photo, and choose your engine. Nano Banana is the one for trend colour and festival glow; GPT Image 2.5 handles fine skin, jewellery and fabric detail more cleanly. Both cost 3 credits per edit, render at 1K and finish in about 40 seconds. Failed generations refund their credits automatically, so experimenting is cheap.

Step 4: read the render like an editor

If the face drifted, the upload photo was too dark or too filtered. If the outfit ignored your prompt, the nouns were probably buried in one long sentence; split the prompt so each of the six parts gets its own clause. If the mood fell flat, you likely described the scene without naming the light. Nine times out of ten, adding two lighting words fixes more than rewriting everything.

A quick word on cost. Every Nano Banana render on this site is 3 credits at 1K, the same as GPT Image 2.5, so engine choice never changes the bill. New accounts get 6 credits, which is two full edits: enough to run the same prompt through both engines once and see which finish you prefer on your own face.

For look ideas, see the Diwali page and the girl guide. For the trend itself, what prompt seen means covers the backstory. Then make your first edit; new accounts get two free.