Every photo editor has two lists: what it will change, and what it will hold still. Nano Banana Pro’s documentation publishes both, which turns most of the work into copying the right shape of sentence.
Nano Banana Pro as a photo editor: the short answer
Describe the change in one sentence, then describe what must not move in the same sentence. Google’s own examples do exactly that: “Turn this scene into nighttime” on one side, “keep the can exactly the same” on the other.
Write only the first half and the model is free to redraw everything else.
The rule Google publishes: be direct and specific
The prompting guide states it without hedging: “For modifying an existing image, be direct and specific. (e.g., change the man’s tie to green, remove the car in the background)”. Two verbs, two nouns, no adjectives about mood.
When the edit depends on uploaded references, the same page adds a second rule: “clearly define the role of each. (e.g., “Use Image A for the character’s pose, Image B for the art style, and Image C for the background environment.”)” Naming which file does what is the difference between an edit and a remix.
Separate what changes from what stays
Google’s model page lists the moves it supports as edits: adjusting colour grades and lighting directions, changing a shot from day to night, altering depth of field to shift the focus, and switching aspect ratio. Around those it publishes prompts built on a lock clause, worded the same way each time: “keep the can exactly the same”, “The character, remains exactly locked in its current position”.
So the working shape has two clauses. Change X. Keep Y.
Anything left out of Y is fair game.
What the official limitations say about editing
Neither page pretends the edit is lossless. The model page writes that masked editing, major lighting changes and blending multiple images “may sometimes produce unnatural results, visual artifacts, or disjointed scenes”. Small faces, accurate spelling and fine details are recorded as ongoing weak spots. The guide repeats the point for editing specifically: “Advanced editing tasks like blending or lighting changes can sometimes produce unnatural artifacts.”
Plan accordingly. One meaningful change per pass, then look at the frame before asking for the next.
Small steps also make it obvious which instruction produced a bad frame.
Worked examples: three edits you can copy
Three edits, each with the instruction text Google publishes and the scene it assumes.
What comes back is described one by one.
Example one: daylight to night
Starting scene: a landscape photo shot in flat afternoon light, sky still bright, no practical light sources in frame.
Instruction: “Turn this scene into nighttime”
What comes back: the same composition with the sky darkened and a night key. Because the instruction says nothing about the subject, nothing is protected — which is the failure risk Google names for day-to-night changes. Add a lock clause when a specific element cannot move.
Example two: moving the focus inside the photo
Starting scene: a portrait where a person stands behind a bank of foreground flowers, everything more or less equally sharp.
Instruction: “Focus on the flowers”
What comes back: the same frame with depth of field pushed to the foreground, flowers crisp and background softened. It is one of the moves the model page lists, altering depth of field to focus on different subjects. Note the economy: two words carried the whole edit.
Example three: translating the text inside the picture
Starting scene: a product shot of three yellow and blue cans with English copy on the label.
Instruction: “translate all the English text on the three yellow and blue cans into Korean, while keeping everything else the same”
What comes back: the same cans with the label copy in Korean, layout and style intact — the developer post describes this as keeping “original artistic style or layout”. Expect slips in grammar and nuance, because translation is listed as a weak area. Have a speaker read the result.
Editing text and other languages inside a photo
Text is where editing and correctness meet. The guide’s advice is to state what the text is and how it should look: “The headline “URBAN EXPLORER” rendered in bold, white, sans-serif font at the top.” The developer post describes the commercial version of the same idea, with menus, signs and documents re-lettered through image-to-image generation while the artwork stays put.
Treat every rendered word as suspect until read. The model page lists accurate spelling as a standing weakness, and the guide repeats it.
The prompt skeleton worth reusing
Four slots, in this order. The change, stated as an imperative. The lock, naming what must not move.
Then the camera or light, when the shot depends on it — the guide suggests directing it like a cinematographer: “A low-angle shot with a shallow depth of field (f/1.8)”, “Golden hour backlighting creating long shadows”. And the canvas, when the output format matters: “A 9:16 vertical poster”, “A cinematic 21:9 wide shot”.
That shape comes straight from the published examples. The Nano Banana Pro prompt guide covers the rest of the framework and the how-to guide covers where the editor lives. For the two generations you may be choosing between, the Nano Banana Pro vs Nano Banana comparison has the table. Builders who want the same edits from code should start with the API guide.
Frequently asked questions
Can it edit a photo I already have, rather than generate a new one? Yes. Google describes Nano Banana Pro as an image generation and editing model, and the guide gives instructions for modifying an existing image.
What is the single most useful rule? Be direct and specific, and say in the same sentence what must stay unchanged.
Can it translate text inside a photo? Yes, and it preserves layout and style. Grammar and cultural nuance can slip.
Does it keep a person looking like themselves across edits? The model page supports up to five characters, and calls character consistency reliable but not guaranteed.
Can I change the aspect ratio without redrawing the subject? Yes. The model page shows the wording: “change aspect ratio to 1:1 by reducing background. The character, remains exactly locked in its current position”.
How much does an edit cost? Not published on the pages reviewed here, so no figure is quoted.
Write the lock, not just the change.