Loading...
A reference image rarely explains the entire edit. On Nana Banana AI, pair your images with instructions about the subject to preserve, the area to replace, and the mood to borrow. Use the generator above to turn that brief into one image you can inspect before your next revision.
A useful editing brief separates the material you want to keep from the changes you want to explore.
Use image 1 for the main subject. Identify the shape, pose or arrangement you would like retained, rather than assuming all details will carry over.
Specify a background, object, palette or lighting adjustment. Keep unrelated instructions out of the first draft so that the result is easier to review.
Describe image 2 as a setting or image 3 as a color guide. Up to 16 references can inform one output; more references do not mean more outputs.
GPT Image 2.5 on Nana Banana AI supports text prompts and image references through Flare and Sunburst. Both versions offer the same input limit and resolution choices. Start with Flare, the default, or select Sunburst before submitting a new request.
For an editing task, choose Image to Image and upload the references in the order used in your prompt. Every request still needs text. If an image contains elements you do not want copied, say which details serve only as inspiration. Review the generated details before using the result.
List the key subject, framing and identifying details that matter to the brief.
Describe the intended replacement or combination in concrete visual terms.
Keep a small checklist alongside each generation so you can identify the next useful change.
Check shape, placement and small details before judging the background or color treatment.
Use a fixed aspect ratio when the image has a destination. Auto is available when you want the model to choose the composition.
1K costs 3 credits, 2K costs 5, and 4K costs 7. Both versions have the same price and references add no extra charge.
These are starting prompts, not generated-result claims. Upload your own images and adapt the details.
Upload a photo of a blue ceramic mug as image 1. Select 1K and 4:3. Prompt: "Use the blue mug in image 1 as the main subject. Place it on a pale wooden kitchen table beside a folded linen towel. Keep the handle visible on the right. Replace the original background with a softly lit wall, and leave the mug free of new lettering." Inspect the handle and silhouette before requesting another version.
Upload a chair photo as image 1 and a room reference as image 2. Choose 2K and 3:2. Prompt: "Use the chair in image 1 as the focal object. Borrow the warm neutral palette and window-light direction from image 2. Place the chair near a plain plaster wall with space around its legs. Do not add the other furniture from image 2." Compare the object, setting and shadows separately.
Use one round to answer one visual question.
Choose the smallest useful set of images and upload them in the sequence described in the prompt.
Write what to keep, change and combine. Choose Flare or Sunburst, then the resolution and aspect ratio; confirm the displayed credits.
Generate one image, check the details, and download it if useful. You may upload that result as a reference in a new paid request. My Creations keeps the task available.
Input roles, output options and task handling.
Yes. Text to Image starts from a written brief. Image to Image adds visual context, but a nonempty prompt is required in either mode.
No. The limit is sixteen input images, and each request produces one output image. Give each input a distinct role rather than adding near-duplicates.
Review the result against your original image. A prompt can request preservation, but generated details may vary. This workflow does not provide a pixel-lock or mask tool.
The selected version is sent with your next request. Flare is the default; both versions support the same resolutions, references and credit prices. Evaluate the output against your own brief.
Choose 1K, 2K or 4K, with Auto or one of ten aspect ratios: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 5:4, 4:5 and 21:9. Custom pixel dimensions are not an input setting.
A confirmed failed generation receives a full refund of its generation credits. A slow response or unavailable status check can still be retried from the task; it does not by itself confirm failure. Submitted tasks cannot be cancelled.
Choose your references, identify their roles, and make one focused edit. The generator shows the required credits before submission.