Nano Banana 2 image editing tutorial (2027)

This guide was checked against Google’s documentation on September 13, 2026. Google’s model lineup, app access, quotas and pricing can change, so use the live model selector and API documentation before a production run. The API model lineup and lifecycle below were rechecked October 10, 2026; that check does not establish the model selected in every Gemini Apps account.

Nano Banana model names explained

Google’s current image-generation guide identifies five Nano Banana API models and recommends Nano Banana 2.1 for new projects. Google launched Nano Banana 2 as Gemini 3.1 Flash Image on February 26, 2026. All images generated by the documented Gemini image models include SynthID.

How to edit an image with Nano Banana

1. Choose the model and surface. In the Gemini app, use a current model that exposes image generation; in the API, select the exact documented image model ID.

2. Upload the source. Use an image you are allowed to process and avoid unnecessary personal or confidential material.

3. Request one bounded change. State the edit, its location and the properties that must remain fixed.

4. Generate and inspect the entire frame. Check the target area plus faces, hands, text, logos, edges, geometry, reflections and background.

5. Continue from an approved version. Save the good result before asking for another change; do not stack edits on an unnoticed error.

6. Export and document. Record the model, prompt, reference images, selected output, human edits, rights and date.

A prompt structure that makes edits easier to review

Use four parts: the change, the location, the preservation constraints and the output requirement.

Workflow recipe

Create a high-definition close-up photographic portrait of the subject, an adult Asian person with short pink hair, wearing a white shirt with the collarbone subtly visible, a tattoo on the left side of the neck, and silver necklace and earrings. The subject extends the right hand toward the camera with the palm facing the lens and fingers slightly curved, creating a natural foreground perspective. Frame the face and upper body prominently, positioning the subject slightly right of center, with a soft blue sky and white clouds in the background. Use shallow depth of field with gentle background blur, realistic natural lighting, fine skin and fabric texture, and a dreamy pastel palette of soft pink and blue. Photorealistic, clean composition, balanced three-quarter framing, natural proportions and authentic photographic detail.

Model

gpt-image-2

Format

9:16

Copy settingsTry in AI Image Editor

A detailed preservation list does not guarantee perfect fidelity. Its value is that it gives the reviewer an explicit checklist and makes the next correction more precise.

Useful edit patterns

• Product cleanup: remove a small background distraction while preserving the product silhouette, label, color and reflections.

• Background replacement: specify the new environment and preserve the subject, camera angle, crop, lighting direction and contact shadow.

• Wardrobe or color change: identify the exact garment or object and preserve skin, pose, fabric boundaries and the rest of the palette.

• Multi-reference composition: assign each uploaded image a role and state which identity, object, style or layout should come from each one.

• Text inside images: provide the exact copy, hierarchy and placement, then proof every character at full resolution.

• Style exploration: describe materials, medium, era, palette, light and composition instead of asking to copy a living artist.

How to edit an image in the Gemini app

Google’s current Gemini Apps help page says users can generate an image, upload an image to edit, or upload multiple images and ask for a new image based on them. In the current workflow, open the image in the Gemini library or upload it, choose Chat, enter the edit request and submit.

Availability varies by account, age, language, country and plan. Google says image editing in Gemini Apps is unavailable to users under 18 and tells users to respect copyright and privacy rights before relying on, publishing or using generated content.

How to use Nano Banana through the API

For a new API workflow, Google’s current guide selects gemini-nano-banana-2.1. Use the exact documented model ID, pass text and image inputs through its supported request, and retain the response with the prompt and source references. Do not use “Nano Banana” as an ambiguous model identifier in code or cost reports.

• Prototype with representative inputs. Include the number and type of references used in production.

• Separate retries from approvals. A successful response is not an approved asset.

• Track cost by exact model. Nano Banana 2.1, 2 Lite, 2, Pro and the legacy model have different roles and pricing.

• Preserve provenance. Keep source asset IDs, prompt, model ID, date and human edits with the delivered file.

• Handle migration explicitly. Google’s Gemini API deprecation table lists March 15, 2027 as the earliest shutdown date for stable gemini-2.5-flash-image and recommends gemini-3.1-flash-lite-image. The stable Nano Banana 2 endpoint has no shutdown date announced, but its recommended replacement is Nano Banana 2.1. This is Gemini API guidance; check other providers separately.

Common failures and fixes

• The subject changed: return to the last approved source and narrow the edit area and preservation list.

• Text is misspelled: supply exact copy and placement; proof the output character by character or add final typography in a deterministic editor.

• Several references were blended incorrectly: number the references and describe one responsibility for each.

• Repeated edits degraded the image: restart from the best approved version instead of continuing the damaged branch.

• A product detail was invented: replace it with a verified source image or correct it manually before commercial use.

• Output dimensions are wrong: state the aspect ratio and delivery size, then verify the downloaded file rather than the preview.

When to use Magic Hour instead

Use Magic Hour’s AI Image Editor when you want a guided edit inside the broader Magic Hour workflow, or the AI Image Generator when the job starts from a prompt rather than an existing source. After approving the still image, use Image-to-Video when the deliverable needs motion.