Nano Banana 2.1: photo editing, prompts and what changed

What changed in Nano Banana 2.1?

Google describes improvements in image realism, instruction following, text rendering and consistency across successive edits. The API documentation lists image outputs at 1K, 2K and 4K, with 1K as the default. It also describes combining multiple reference images.

These capabilities do not remove the need for review. Google’s model card lists remaining problems with small text, long passages, character consistency and partial instruction following. Treat an attractive first draft as something to inspect, especially when a label or product shape matters.

Sources: Google AI for Developers: Gemini Nano Banana 2.1 · Google DeepMind: Nano Banana 2.1 model card

Nano Banana 2.1 vs Pro: versions and free access

Google’s consumer image page distinguishes free Nano Banana 2 from Nano Banana 2.1 unlocked through a Google AI plan. Check your account’s current plan, region and model selection before paying. App access and developer API usage are different purchase paths; do not assume one subscription covers both.

Nano Banana 2.1 updates Nano Banana 2. The developer documentation presents it as the efficient counterpart to Nano Banana Pro, rather than another name for Pro. When comparing results, write down the exact model displayed, output size and interface so that you know what you actually tested.

Sources: Google Gemini: image generation and plan access · Google AI for Developers: Gemini Nano Banana 2.1

Product photo prompt: change the setting, preserve the cup

For a real product listing, start with a clear photo of the actual item. A description alone invites the generator to invent its shape. Choose a reference that shows the rim, handle, glaze and base, and keep an unedited copy beside your result.

This exercise changes the setting while keeping the product fixed. Inspect the handle opening, thickness of the rim and glaze marks at full size. If the cup becomes taller or smoother, reject that version even if the background looks more expensive.

Edit my uploaded ceramic cup photo for a product listing. Preserve the cup’s exact silhouette, handle, rim, glaze pattern and color. Place it on a pale stone surface with a plain warm-gray background and soft window light from the left. Keep one cup only, fully visible, in a square composition. Add a natural contact shadow. Do not add a logo, text, steam or accessories. If a detail is not visible in the reference, do not invent a decorative feature.

Text in images: a short menu prompt you can check

Give the generator a small amount of final copy instead of asking it to write and design everything together. For this fictional café menu, decide the words first. You can replace them with your own language, but keep a separate text file as the reference.

Check every letter at the final display size, including accents and punctuation. For a real menu, also verify prices and currency yourself. If repeated attempts distort a line, generate the background without lettering and add the approved text in a design editor.

Create a simple square menu graphic for a fictional café. Use a cream background, dark brown lettering and one small hand-drawn cup in a corner. Include exactly these three lines: “Today’s menu”, “Filter coffee”, “Oat latte”. Keep the text large, evenly spaced and centered. Do not add prices, extra words, logos or decorative lettering. Leave generous margins.

Consistent image editing: make one change at a time

Once you approve a version, save it before another edit. Give the next request a clear boundary: what may change, and what must stay. For a series of product images, use the same approved reference rather than a chain of increasingly altered copies.

The follow-up below asks only for space around the cup. Compare the original and edited versions side by side, not from memory. When a detail drifts, return to the approved reference and describe that specific mismatch. Repeating “make it better” gives you no useful acceptance criterion.

Use the approved cup image as the reference. Expand the canvas to a horizontal 16:9 composition, keeping the cup on the right and leaving the left half empty for text that I will add later. Preserve the cup’s shape, color, glaze, lighting and apparent size. Continue the existing background naturally. Do not generate text or add objects.

A practical checklist before you publish an AI image

Our suggested workflow is to judge accuracy first, then presentation. Give each item below a pass or fail and keep the rejected files long enough to identify recurring mistakes. This is a quality-control method, not a performance ranking of image models.

If you compare two tools, give both the same reference and request. Record the number of revisions and the time spent fixing the result. The useful winner is the one that gets your own job to an acceptable finish with less correction, not necessarily the most dramatic first image.

• Identity: does the object still match the reference, including edges and small marks?
• Text: are every word, number and symbol correct and readable at the intended size?
• Composition: are the crop, empty space and shadow suitable for the destination?
• Delivery: have you saved the approved version and checked the exported file?