Nano Banana Product Photography: 8 Prompts Tested

For Nano Banana product photography, edit a real photo of your product instead of describing it, and tell the prompt to keep the label unchanged. We ran 8 prompts on Nano Banana 2.1 and Nano Banana Pro on November 2, 2026: 13 of the 14 labeled images kept every word, and naming the words fixed the one miss.

Short answer

• Prompt order: the job, what stays the same, background, light, framing, then any exact text.
• Which model: neither won outright, so start on Nano Banana 2.1, which Google lists at $0.0504 per 2K image against $0.134 for Pro (checked November 2, 2026).
• Labels and details: when water, hands or props come near the product, write the exact words, direction and color into the prompt.
• Before you publish: compare each image with the real product, because our runs mirrored a shoe and added an extra coffee filter.

How to write a Nano Banana product photography prompt

Start from a real photo of your product and write the prompt in the order below. Google's product-shot template names the product, what it sits on and the lighting (image generation guide, checked November 2, 2026), and for edits it says to describe in detail what must not change.

The 85% frame height is our own framing choice, and no marketplace rule set it.

How we tested

Each prompt ran as an edit of a source image on both models at 2K through the OmniaKey API on November 2, 2026. Prompt 6 ran twice per model, so 18 edits, and none failed; the median call took 41.9 seconds on 2.1 and 46.0 on Pro.

The source photos look like phone photos, but GPT Image 2 made them, and AURELLE and HILLTOP ROAST are invented brands. We checked every label by eye at full size. These are 1 or 2 runs per prompt on generated photos, so results on your own photo may differ.

The times printed on the images are how long each call took.

8 Nano Banana product photography prompts and what came back

Each recipe is the prompt we sent and what we saw at full size.

1. White background main image

The main image for a shop listing: the bottle on clean white.

Turn this phone photo into a clean e-commerce main image. Keep the bottle, the dropper and every word on the label exactly as they are. Place it on a pure white background (#FFFFFF) with only a soft, natural contact shadow under the base. Even studio lighting from the front and both sides, no reflections of the kitchen in the glass. Center the bottle, straight on, filling about 85% of the frame height.

Settings: edit, 1:1, 2K, serum photo. Nano Banana 2.1 32.5 s, Pro 38.2 s.

2.1 returned all 7 sampled background patches at exactly #FFFFFF. Pro's was light gray, #F0EEF1 top left to #FCFBFB bottom right; labels exact.

Tip: on Pro, ask for "solid pure white (#FFFFFF) from edge to edge, with no gradient".

2. Studio hero shot with room for a headline

A hero image with room for a headline.

Use the serum bottle in this photo as the product for a premium studio hero shot. The bottle stands on a terracotta-colored stone block in front of a warm sand-colored seamless backdrop. A single hard key light from the upper left casts a crisp diagonal shadow to the right, and a thin rim light traces the edge of the amber glass. Keep the label text "AURELLE", "Vitamin C Serum" and "30 ml" unchanged. Leave clear empty space above the bottle for a headline.

Settings: edit, 4:5, 2K, serum photo. Nano Banana 2.1 35.3 s, Pro 44.8 s.

Both delivered the terracotta block, sand backdrop, diagonal shadow and empty space above; labels exact.

3. Lifestyle shot on a bathroom shelf

A lifestyle shot for a product page.

Place this exact serum bottle on a white marble bathroom shelf next to a folded oatmeal linen towel and a small sprig of eucalyptus. Soft morning sunlight comes through a frosted window on the right and casts gentle leaf shadows across the shelf. Shallow depth of field: the bottle is the sharpest thing in the frame, with the label facing the camera and fully readable. Photorealistic, shot like an 85 mm lens at eye level.

Settings: edit, 4:5, 2K, serum photo. Nano Banana 2.1 82.9 s, Pro 43.3 s.

Both put the bottle on the marble shelf beside the towel and eucalyptus in soft window light, labels readable. The 82.9-second call logged one network error and still finished.

4. Overhead flat lay

An overhead flat lay with coffee props.

Create an overhead flat lay for a coffee shop's product page using this coffee bag. Lay the bag flat in the center of a light oak table, surrounded by a ceramic pour-over dripper, a small pile of roasted coffee beans, a white cup of black coffee and a folded paper filter. Soft daylight from the top of the frame, gentle shadows, balanced spacing, nothing touching the edges. Keep the label design and all of its text exactly as in the photo.

Settings: edit, 1:1, 2K, coffee bag photo. Nano Banana 2.1 45.1 s, Pro 48.4 s.

Labels exact on both. 2.1 drew a proper ceramic dripper but two paper filters; Pro's looks like a mug holding a filter, with more open space.

5. On a runner's foot

An in-use shot for a sports ad.

Show this exact sneaker being worn by a runner mid-stride on a red running track at golden hour. Crop from the knee down with the camera low near the ground. The shoe is in sharp focus and keeps the white upper, the mint-green sole, laces and heel tab from the photo; the track lines behind are blurred with motion. Photorealistic sports campaign look, no logos added.

Settings: edit, 3:2, 2K, sneaker photo. Nano Banana 2.1 29.8 s, Pro 50.8 s.

Colors right on both, no logos added. 2.1 showed two feet from the front with dust rising; Pro showed one foot from the side with motion blur, keeping the mint heel tab and staying closer to the source.

Tip: if the heel has to match your photo, try Pro.

6. Social ad with text

A square social ad with a headline and subline.

Make a square social media ad for this coffee bag. Put the bag on the right side of the frame on a deep forest-green background with a few scattered coffee beans. On the left, set the headline "Roasted every Monday" in a bold cream serif font, and under it, in smaller cream sans-serif text, "Free shipping over $30". Keep the bag's own label unchanged. Clean layout with generous margins and no other text.

Settings: edit, 1:1, 2K, coffee bag photo. Nano Banana 2.1 46.5 and 41.9 s, Pro 33.5 and 46.0 s.

All 4 images have the headline, subline and label correct. Pro run 1 broke the subline across two lines; 2.1 run 2 turned the floor charcoal navy instead of forest green.

Tip: Google's prompting tips say to state the exact text and how it should look.

7. Three colorways from one shoe

A three-colorway lineup for a product page.

Using this sneaker as the base design, create a product lineup of three colorways side by side on a light gray studio background: on the left the original white with mint-green sole and laces, in the middle an all-black version with a white sole, on the right a sand-beige version with a brown sole. Same side-profile angle and same size for all three, evenly spaced, soft shadows under each. Keep the shape, stitching and sole pattern identical to the photo.

Settings: edit, 16:9, 2K, sneaker photo. Nano Banana 2.1 76.1 s, Pro 66.6 s.

Colors right on both. 2.1 kept the toes pointing right and the embossed mark; Pro mirrored all three and dropped the mark.

Tip: add "every toe pointing right exactly as in the photo" and "Do not mirror the shoe".

8. Splash shot

An ad shot with a water splash.

Create a bold advertising shot of this serum bottle: it stands in a shallow pool of clear water while a crown of water splashes up around it, frozen in mid-air, with two thin orange slices floating nearby. Bright citrus-orange gradient background, high-speed flash lighting, every droplet crisp. Keep the bottle, the dropper and the label exactly as in the photo.

Settings: edit, 4:5, 2K, serum photo. Nano Banana 2.1 35.8 s, Pro 54.2 s.

2.1 let the water cross the bottom of the label and garbled "30 ml"; its background came out darker than "bright citrus-orange". Pro kept all the label text readable.

Tip: when water can cross the label, name the words in the prompt.

Three failures and the wording that fixed them

Every failure came right when we changed the prompt and kept the same model. Google warns that its models "can still struggle with small faces, accurate spelling, and fine details" (DeepMind's model page), and a user on Google's forum wrote on September 30, 2025 that "the logos are changed regardless of the prompt highlighting to keep the product 'exactly the same'" (forum topic).

The water garbled the label

On 2.1 the water crossed the label's bottom and "30 ml" came out garbled.

Create a bold advertising shot of this serum bottle: it stands in a shallow pool of clear water while a crown of water splashes up around it, frozen in mid-air, with two thin orange slices floating nearby. Bright citrus-orange gradient background, high-speed flash lighting, every droplet crisp. Keep the bottle and the dropper exactly as in the photo. The label stays dry, fully visible above the splash and faces the camera. It reads "AURELLE", "Vitamin C Serum" and "30 ml", with no other text on it.

All three label lines were crisp in 2 of 2 runs (42.0 and 29.2 s), and the background was brighter.

The shoe came back mirrored

On Pro all three shoes came out mirrored, toes left, embossed mark gone.

Using this sneaker as the base design, create a product lineup of three colorways side by side on a light gray studio background: on the left the original white with mint-green sole and laces, in the middle an all-black version with a white sole, on the right a sand-beige version with a brown sole. Same side-profile angle and same size for all three, every toe pointing right exactly as in the photo, evenly spaced, soft shadows under each. Do not mirror the shoe. Keep the shape, stitching, the embossed mark on the side of the heel and the sole pattern identical to the photo.

All three toes point right, and the mint heel tab and embossed mark stayed (32.0 s); the source's lace bow became neat lacing.

The white background came out gray

On Pro the background was a light gray gradient, from #F0EEF1 to #FCFBFB.

Turn this phone photo into a clean e-commerce main image. Keep the bottle, the dropper and every word on the label exactly as they are. The background is solid pure white (#FFFFFF) from edge to edge, with no gradient, no vignette and no gray in the corners. The only shadow is a soft, natural contact shadow directly under the base. Even studio lighting from the front and both sides, no reflections of the kitchen in the glass. Center the bottle, straight on, filling about 85% of the frame height.

All 7 sampled patches were #FFFFFF, against #F0EEF1–#FCFBFB before, and the label is exact (31.6 s).

All three fixes name the thing to keep, in words: the label text, the toe direction, the exact color. The splash fix held in both of its runs, and the other two fixes ran once each. A user on the same forum reported blurry boxes on white backgrounds at 4K in June 2026 (topic 171932), so zoom in on any 4K white background before you use it.

Which is better for product photography: Nano Banana 2.1 or Pro?

Neither model won outright: 2 wins each, 3 ties and 1 mixed result over 8 prompts.

Google's list price (checked November 2, 2026) is $0.0504 per 2K image for 2.1 and $0.134 for Pro; Pro costs the same at 1K and 2K, so only 4K costs more. Median call: 41.9 seconds on 2.1, 46.0 on Pro.

Start on Nano Banana 2.1: it costs less, and on our runs it won the white background and the colorways. When a shot fails, change the wording before the model, which fixed all three failures.

For a wider comparison outside product shots, see Nano Banana 2.1 vs Pro. For OmniaKey's current rates see the Nano Banana 2.1 and Nano Banana Pro pages.

Run these prompts through the API

To skip the code, use the AI image generator with your photo as a reference image. Through the API, the flow in OmniaKey's media API docs is create, poll, download: POST /v1/media/tasks returns a task id, then GET /v1/media/tasks/{id} reports progress until it finishes. The body carries model, type, operation, input and parameters, and the request needs an Idempotency-Key header.

"""Edit one product photo with Nano Banana through the OmniaKey media API.

Usage:
    export OMNIAKEY_API_KEY="your-omniakey-api-key"
    python3 nb_product_edit.py photo.jpg prompt.txt product.jpg

The photo is a JPEG under 700 KB, sent inline as Base64.
"""
import base64
import json
import os
import sys
import time
import urllib.error
import urllib.request
import uuid

API = "https://api.omniakey.com/v1/media/tasks"
MODEL = "gemini-nano-banana-2.1"  # or "gemini-3-pro-image" for Nano Banana Pro
HEADERS = {
    "Authorization": "Bearer " + os.environ["OMNIAKEY_API_KEY"],
    "Content-Type": "application/json",
    "User-Agent": "nano-banana-product-example/1.0",
}
POLL_SECONDS = 5
GIVE_UP_AFTER = 5 * 60  # most of our edits finished in under 90 seconds


def call(method, url, body=None, extra_headers=None):
    data = json.dumps(body).encode() if body is not None else None
    req = urllib.request.Request(url, data=data, method=method, headers={**HEADERS, **(extra_headers or {})})
    for attempt in range(3):
        try:
            with urllib.request.urlopen(req, timeout=60) as resp:
                return json.load(resp)
        except urllib.error.HTTPError as err:
            sys.exit(f"HTTP {err.code}: {err.read().decode(errors='replace')}")
        except (urllib.error.URLError, TimeoutError, ConnectionError) as err:
            if attempt == 2:
                raise
            print("network error, retrying:", err)  # a retried POST keeps its Idempotency-Key
            time.sleep(5)


def main(photo, prompt_file, out_path):
    raw = open(photo, "rb").read()
    if len(raw) > 700_000:  # Base64 adds a third, and the whole request must stay under 1 MiB
        sys.exit("The photo is over 700 KB. Resize it first, for example to 1024 px on the long side.")
    task = call("POST", API, {
        "model": MODEL,
        "type": "image",
        "operation": "edit",
        "input": {
            "prompt": open(prompt_file, encoding="utf-8").read().strip(),
            "image": {"data": base64.b64encode(raw).decode(), "mime_type": "image/jpeg"},
        },
        "parameters": {"aspect_ratio": "1:1", "image_size": "2K"},
    }, {"Idempotency-Key": str(uuid.uuid4())})
    print("task", task["id"], task["status"])

    start = time.time()
    while task["status"] in ("queued", "processing"):
        if time.time() - start > GIVE_UP_AFTER:
            sys.exit(f"Still {task['status']} after 5 minutes. Do not submit it again; "
                     f"check it later with GET {API}/{task['id']}")
        time.sleep(POLL_SECONDS)
        task = call("GET", f"{API}/{task['id']}")
    print(task["status"], "after", round(time.time() - start), "s")
    if task["status"] != "succeeded":
        sys.exit(json.dumps(task.get("error") or task))

    # The signed download URL is short-lived, so save the file right away.
    asset = task["assets"][0]
    req = urllib.request.Request(asset["url"], headers={"User-Agent": HEADERS["User-Agent"]})
    with urllib.request.urlopen(req, timeout=120) as resp, open(out_path, "wb") as f:
        f.write(resp.read())
    print("saved", out_path, asset.get("mime_type"))


if __name__ == "__main__":
    main(*sys.argv[1:4])

export OMNIAKEY_API_KEY="your-omniakey-api-key"
python3 nb_product_edit.py photo.jpg prompt.txt product.jpg

task task_UdWluqJc0prstORG3z2PpK51OhMrW3Oc queued
succeeded after 25 s
saved data/t1-check.jpg image/jpeg

• One run, white background prompt on 2.1: a 158,554-byte, 1024 px JPEG in, a 2048×2048 JPEG of 739,044 bytes out.
• The whole request has to fit in 1 MiB including the Base64 photo, which is why the script stops at 700 KB.
• Change aspect_ratio and MODEL for other jobs: our runs used 1:1, 4:5, 3:2 and 16:9. Google's Nano Banana 2.1 model page lists 4:1 and 8:1 banners, and Pro's 10 ratios include neither.
• Google's models usually return JPEG; the real format is in assets[].mime_type.
• This file ran once; the other edits used our batch tool on the same API.

Frequently asked questions

Can Nano Banana keep my logo and label text?

Usually, but not always: 13 of the 14 labeled images kept every word, and naming the words fixed the one miss. Google's edit advice is to describe a logo in great detail along with the edit request, and its limitation note warns about spelling and fine details. So write the label words into the prompt and check every image at full size.

Do Nano Banana product photos have a watermark?

Every image from the API carries an invisible SynthID mark, so you cannot see a watermark. Google said at the Pro launch in November 2025 that the visible Gemini sparkle stays on images made by free and Google AI Pro users in the Gemini app, and is removed for AI Ultra subscribers and in Google AI Studio (launch post).

Can I use these images in my store?

Google's Gemini API terms say it does not claim ownership of generated images, but it may generate the same or similar content for others, and you are responsible for how you use them. Its Prohibited Use Policy bars claiming an image was made solely by a human in order to deceive, and using personal data or biometrics without legally required consent. Compare each image with the real product too: our runs mirrored a shoe and added an extra coffee filter.

How many reference images can I use?

Google allows up to 14 reference images in one request, split per model: 10 objects and 4 characters on 2.1, or 6 objects, 5 characters and 3 style references on Pro. OmniaKey caps the whole request at 1 MiB, Base64 images included, so several references must each be small.

Is Nano Banana 2 still usable?

Yes, for now: Google deprecated Nano Banana 2 (gemini-3.1-flash-image) on October 6, 2026, gave no shutdown date and says to move to Nano Banana 2.1 (changelog). Its list prices are still on Google's pricing page (checked November 2, 2026).

Sources checked

• Image generation with Gemini, checked November 2, 2026
• Gemini API changelog, checked November 2, 2026
• Nano Banana 2.1 model page, checked November 2, 2026
• Gemini API pricing, checked November 2, 2026
• Nano Banana Pro launch post, checked November 2, 2026
• Gemini Image Pro, DeepMind, checked November 2, 2026
• Prompting tips for Nano Banana Pro, checked November 2, 2026
• Gemini API additional terms, checked November 2, 2026
• Generative AI Prohibited Use Policy, checked November 2, 2026
• Forum: white background thread, checked November 2, 2026
• Forum: logos changed thread, checked November 2, 2026

OmniaKey's own documentation, checked November 2, 2026: image API and media API overview.