Editing photos in Nano Banana with plain text

The Nano Banana models

Google ships several models under one name. They differ in how many references they take and in resolution. The figures below come from the Gemini API documentation.

How to phrase an edit

A three part formula works best: what to change, what it should look like, and what stays the same. Describe the result as a full sentence rather than a pile of keywords. Describe what you want instead of what you do not: "an empty street" is clearer than "no people".

• Background: "Replace the background with a bright studio and a soft shadow. Keep the person, pose, clothing and the light on the face unchanged"
• Removal: "Remove the wires above the house and fill in the sky. Leave the rest of the photo as it is"
• Outfit: "Dress the woman in a blue linen dress. Face, hair, pose and background stay the same"
• Light: "Add warm evening light from a window on the left, like at sunset. Do not change the composition"
• Style: "Redraw the photo as a watercolor while keeping the composition and facial features"

Keeping a face consistent

Nano Banana 2, 2.1 and Pro accept dedicated character images. Upload two or three sharp photos of the face and state plainly that the appearance comes from them. The more precise the instruction, the less the model alters the features.

Over a long chain of edits, features slowly drift. Google suggests starting over with a detailed description of the person when that happens. Make one edit per step and save the good intermediate versions.

Keeping products and labels intact

Name the object and list what matters about it: shape, color, material, logo, label. "Keep everything else" works worse than a list of concrete details. To change text, give the exact words in quotes along with the font and position.

Check the label letter by letter after each edit. If the lettering must stay perfect, place the original product over the new background in the design editor.

Working with several photos

Nano Banana can build a new scene from several images. Give each one a role: "photo 1 is the person, photo 2 is the jacket, photo 3 is the room". Without roles the model decides on its own what to take from where.

Edits usually keep the aspect ratio of the input. With several inputs the ratio may come from the last image, so state the format you want explicitly.

Common mistakes

When the result is off, simplify the request before you add to it.

• Several edits in one sentence. The model does only part of them or changes extra things
• A vague goal. "Make it nicer" gives the model nothing concrete to change
• No list of what stays. The model may change the face or background along the way
• A face that is too small in the source. Features get reconstructed inaccurately

How to do it in Solistra

The Image section offers Nano Banana, Nano Banana 2, 2.1, 2 Lite and Pro. You can send the result straight to video or to the design editor.

• Open Image and choose a Nano Banana model
• Upload the source photo and any references
• Write the edit as what to change, how, and what to keep
• Check the credit price on the button and run it