Nano Banana Pro Image-to-Image
Edit photos with Nano Banana Pro on Conut.ai: up to 14 reference images, 1K–4K output, ten aspect ratios and 30–60 credits by resolution. Prompts, limits and examples.
Nano Banana Pro Edit is the image-to-image side of Google's Nano Banana Pro model. Upload one photo or a set of references, describe the change in plain language, and the model returns a new image at 1K, 2K or 4K. The page accepts up to 14 images, which makes it useful for multi-reference work: hold a character's face from one file while taking the outfit from another, or place a product into a scene without redrawing it by hand. An output costs 30 credits at 1K, 40 at 2K or 60 at 4K.
- Resolution: 1K / 2K / 4K
- References: Up to 14
- Price: 30/40/60 by resolution
- Output: PNG / JPEG
How to use Nano Banana Pro Edit
Start with identity and composition, then make the smallest change that achieves the brief.
- Upload the source and references Put the image to edit first, then add supporting references for faces, outfits, objects or style. Up to 14 files, 10 MB each.
- Write the change request Name the region to edit and the result you expect. Add a keep-unchanged clause for face, pose, background and framing.
- Draft at 1K Use 1K while testing the instruction. It costs less and returns faster, which makes it practical to compare two or three prompt variants.
- Set the final resolution and ratio Move to 2K or 4K for the winning version. Choose the aspect ratio before the final run so the composition is not cropped afterward.
- Generate and inspect at full size Check the edited region against its surroundings, then check hands, text and fine patterns. Keep the PNG if another edit will follow.
Settings reference
The Pro edit page adds a resolution tier and multi-reference input to the standard edit flow.
| Setting | Value | Notes |
|---|---|---|
| Resolution | 1K / 2K / 4K | Default 2K. 1K is useful for drafts; 4K is the final-output tier. |
| Aspect ratio | 10 options | 1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9 and 21:9; default 3:4. |
| Reference images | Up to 14 | JPG, PNG or WebP files up to 10 MB each. More references help with character sheets and product sets, but each one needs a role in the prompt. |
| Output format | JPEG / PNG | PNG for further editing, JPEG for delivery. |
| Prompt | Edit instruction | State the change, the invariant features and which reference supplies each new element. |
| Price | 30 / 40 / 60 credits | Billed by output resolution: 1K, 2K and 4K respectively. |
What the edit page is for
The model is prompt-driven rather than mask-driven. You do not paint a selection; you describe what should change and what must stay the same. In practice that means naming the subject's identity, pose and framing first, then the edit: swap the jacket, move the scene to a rainy street, replace the label text, match the light to the background.
Because the page accepts several references, it also works as a composition tool. Give each image a role in the prompt — first image is the person, second is the jacket, third is the room — and describe how they combine. The more references you add, the more explicit the roles need to be.
- Background replacement while keeping the subject intact
- Outfit, material and colour changes on a product or person
- Style transfer from a reference image
- Text replacement on packaging, signage and mockups
- Multi-reference scenes with a person, product and setting
- Cleanup and restoration of older photos
Resolutions, references and credits
| Setting | Options | Notes |
|---|---|---|
| Resolution | 1K / 2K / 4K | 1K for drafts, 2K for delivery, 4K for print |
| Reference images | Up to 14 | Assign a role to each image in the prompt |
| Aspect ratio | 1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9 | The output frame, not the input frame |
| Output format | PNG / JPEG | PNG when the edit adds text or graphics |
| Credits (1K / 2K / 4K) | 30 / 40 / 60 | Ultra: 40 at 4K, 50 at 8K |
| Multi variant | 2 images per run | 20 credits for one, 40 for two |
Prompts that keep a face or product consistent
The first sentence is the contract. State what must not change, then state the edit. Vague prompts let the model redesign the whole image, which is the usual cause of “it changed the face”.
- Name the invariant: same face, same pose, same camera angle
- Describe one edit axis at a time: light, then background, then wardrobe
- Quote label text exactly and say where it sits
- For composites, number the references and give each one a job
- Describe the target light direction so the subject matches the new background
- Keep the output aspect ratio close to the input when identity matters
Limits and practical notes
- There is no mask brush; precision comes from wording
- Small text and fine jewellery are the first details to degrade at 1K
- More references can blur roles; three or four well-described images often beat ten
- Each rerun costs 30–60 credits by output resolution, so settle the prompt at 1K before moving to 4K
- For 8K output use the Edit Ultra variant
Prompt examples for Nano Banana Pro Image-to-Image
- Background swap: Keep the woman's face, hair, pose and jacket exactly as in the first image. Replace the background with a quiet Copenhagen street on a rainy afternoon, wet cobblestones, soft grey light, shallow depth of field. Match the light direction on her face to the overcast sky. 4:5.
- Product relight: Keep the sneaker's shape, materials and logo placement from the first image. Place it on a warm sand dune at golden hour, low sun from the right, long soft shadow, fine sand texture, no extra props. 3:2.
- Outfit change: Use the first image as the person: same face, same body, same standing pose. Replace the outfit with an oversized charcoal wool coat and black turtleneck, keep the original studio background and soft frontal light. 3:4.
- Label replacement: Keep the bottle shape, glass reflections and composition from the first image. Replace the label text with "NORTH COAST" in a thin serif, centred, dark green on cream, and reduce the remaining label copy to one small line reading "200 ml". Keep the same lighting. 2:3.
- Style transfer: Keep the house, trees and path from the first image. Repaint the scene in the style of the second image: flat gouache shapes, muted sage and ochre palette, visible paper grain, simplified shadows. 3:2.
- Multi-reference scene: Compose a scene using the first image for the woman's face and identity, the second for the armchair, and the third for the room. She sits in the chair reading, late afternoon light through a tall window, warm wood floor, calm and lived-in. Keep her face and the chair design unchanged. 4:3.
Pro tips for Pro Edit
The extra references and resolution tiers are useful only when the prompt assigns each one a job.
- Draft low, finish high Iterate at 1K, then rerun the winning prompt at 4K. You pay for fewer high-resolution misses.
- Use references as a character sheet Front, profile and three-quarter views give the model more identity evidence than nine near-duplicate selfies.
- Anchor identity explicitly Name the face, hairstyle and distinguishing marks that must survive. The model preserves what the prompt makes explicit.
- Separate composition from styling Fix the pose and framing in one pass, then change wardrobe or color grading in a second. Mixed requests are harder to diagnose.
- Keep the source resolution honest A 4K output from a soft 800 px source will still look soft. Upscaling cannot recover detail the input never had.
Frequently asked questions
How many images can I upload?
Up to 14 references per generation, 10 MB each. One is enough for a simple edit; several are useful for composites.
What resolutions are supported?
1K, 2K and 4K on this page. The Edit Ultra variant adds 8K.
How much does an edit cost?
An output costs 30 credits at 1K, 40 at 2K or 60 at 4K. Ultra costs 40 credits at 4K and 50 at 8K; Multi costs 20 credits for one image and 40 for two.
Does it keep the original face?
It can, but you have to say so. Lead the prompt with “keep the face, pose and framing unchanged” and limit each run to one edit axis.
Is there a mask or lasso tool?
No. Editing is prompt-driven: describe the region and the change in words rather than painting a selection.
Can I use it for product mockups?
Yes. Upload the product and a scene reference, assign each a role, and quote any label text exactly.
What aspect ratio should I pick?
The frame you want to deliver. If identity matters, keep it close to the input image so the model does not crop or extend the subject.
Compare with other image models
- Nano Banana Pro Text to Image
- Nano Banana Pro Edit Ultra
- Nano Banana Pro Edit Multi
- Nano Banana 2 Edit
- Seedream 5.0 Pro Edit
- GPT Image 2 Edit
Specifications
- Provider: Google
- Model type: Image to image
- Credit pricing: 40 credits per image
- Capabilities: starter, high-quality
About Conut.ai
Conut.ai is an AI image, video and audio generation platform with 50+ creative tools. It supports text-to-image, image-to-video, face swap, motion control, relighting, inpainting and video effects through models including Nano Banana Pro, Seedream, GPT Image, Sora 2, Kling, WAN and Seedance. Start free — no credit card required.