Nano Banana AI Photo Editor

Every image below was generated on this page on 18 August 2026, first take, with the exact prompt shown — using the same prompts we ran on the Nano Banana Pro page so you can compare the two directly.

Editing a photo you already have

This is what the model got famous for, and it is where it performed best in our tests. We handed it a product shot generated earlier on our Nano Banana Pro page and asked for a new setting — sunlit windowsill, rosemary beside it, label untouched. It rebuilt the scene and kept the wraparound label intact: the brand line, the em dash, and the small print underneath it. Preserving text that already exists turns out to be a different, easier job for this model than inventing text from nothing — so judge it on the task you actually have.

Display type and layout hold up

Give it a headline and a short supporting line and it sets them correctly. The poster below asked for a heavy condensed headline across the top third, a small subtitle under it, and a photographic scene below — and that is what came back, with the em dash in the subtitle intact. Composition, margins and hierarchy are not where the cheaper model gives ground; in our side-by-side runs the layouts were comparable. What it does not do is match the resolution: our runs came back around 1 megapixel, against roughly 4 on the Pro channel.

Where it gives ground: longer captions and non-Latin script

This is the honest cost of the cheaper model, and it is specific. Asked for a four-stage diagram, it produced a correct layout, correct icons and the right cycle order — and then misspelled it: "EVEPORATE" instead of EVAPORATE, "Assorbs heat freat from outside air", "refirigerant". The Pro channel returned the same diagram with no spelling errors at all. Non-Latin script fared worse: our Korean exhibition title should have been four syllable blocks and came back as five, two of them wrong, while the English subtitle beneath it was fine. If your image needs caption-length text or a script other than Latin, this is the wrong model on this site.

• Anyone editing photos by instruction

  Change a background, a setting, a light source, without masks or layers. This is what the model became known for and the task it handled best in our tests — including keeping existing label text intact.

• People iterating on composition

  Layout, framing and hierarchy came back comparable to the pricier channel at the same prompts. Explore directions here, then re-run the winner somewhere with more resolution.

• Anyone who does not need text in the image

  Photographic and illustrative work with no words in frame sidesteps this model's one real weakness entirely.

• Not for: caption-heavy or non-Latin layouts

  Infographics with several lines of caption, or titles in Korean, Japanese, Arabic or any non-Latin script. Our tests failed on both. Use Nano Banana Pro or GPT Image 2 instead.

Credits, reference-image limits, resolutions and ratio counts are read from our own model registry on 18 August 2026 — the ratio counts exclude "auto", which is a setting rather than a ratio. Speeds and text results are our own measurements with the sample size stated. Cells we have not tested say so rather than borrowing a number from a vendor page.

1. 1

   Open the workspace

   Create a free account — no API key, no setup. The generator on this page is already set to Nano Banana.

2. 2

   Upload, or start from a prompt

   Attach up to five photos and describe the change in plain language, or write a prompt from scratch. Editing is where this model is strongest.

3. 3

   Keep any in-image text short

   A headline and one supporting line is safe. Several lines of caption, or a non-Latin script, is where our tests broke — put those on a different model.

4. 4

   Generate and iterate

   Most runs land in about a minute. History keeps every prompt and parameter, so re-running a variation is one click.