You need three new images this month: one for a Facebook or Instagram post, one for a Google ad, and one for your website. Paying a designer for every variation adds up, and your last batch of stock photos looked like everyone else's. Google's Nano Banana 2.1 image model can turn a short description, or a photo you already have, into a draft image in a few minutes. Using Nano Banana for small business ads comes down to a short routine, covered in the order you do it: choose a tool, start from a real photo, write the prompt, export each size, and check the result before it goes live. It also covers what Google Ads and Meta say about AI-made images. You can make a basic image yourself in an afternoon, and the sections below show where a real photograph is still the better choice.
Key Takeaways
Upload your own product or approved reference photo, then describe the subject, brand colors, setting and ratio you want.
Square, vertical and wide placements each get their own export instead of one image cropped three ways.
Google says generated Ads assets are not guaranteed to pass Ads Policy, and Meta applies its rules whether or not AI made the image.
How do you make your first ad image with Nano Banana?
Pick one real marketing job before you open any tool. A feed post for this week is a better first project than "all our social images." Then work through these steps.
- Name the job: Write one line, such as "square image for a Facebook post announcing our spring booking window."
- Gather your inputs: Your logo, brand colors, a clear photo of the product or service result, and any wording that must appear in the image.
- Pick the tool: Gemini, Google AI Studio or, if your account is eligible, the image tools inside Google Ads.
- Upload the reference photo and write the prompt: State the subject, setting, brand colors, mood and ratio.
- Generate several versions: Compare them side by side and pick the closest, then ask for one specific change at a time.
- Check and save: Review spelling, product shape, claims and labels, then save the final file with the prompt that made it.
Treat the first result as a draft. Google describes the model as a conversational editor, so asking for one change at a time ("make the background a plain light wall") usually serves you better than rewriting the whole prompt.
Which tool should you use to make the image?
You have three no-code places to start, and they differ in cost and in how much control you get.
Google's Gemini page says Nano Banana 2.1 is available through a Google AI Plan, and that the free Gemini experience offers access to Nano Banana features that depend on your account. Google AI Studio is the developer-facing workspace, but you can type prompts and upload images there without writing code. Inside Google Ads, generated images are offered in campaign setup, ad and asset-group editing, the Asset Library and Asset Studio. Google says this feature is gated by account age, policy history, language and vertical, and it is rolling out gradually, so you may not see it yet.
Use Gemini when you want a social or website image. Use the Google Ads tools when the image is going straight into a campaign and your account shows the option. If you do not see the Ads option, make the image in Gemini and upload it as a normal asset.
How do you write a prompt that gets you a usable image?
Google's prompt guide recommends specifying the style, subject, setting, action and composition. For business images, add the product or service, brand colors, lighting, mood, background, platform, aspect ratio, and any required text in quotation marks. Upload clear reference images and give recurring objects or characters a name so the model keeps them consistent. Google Ads adds its own advice: use several angles of the product, one product per reference image, and state the subject, color, style and background.
Copy this fill-in template and swap the brackets:
> Using the attached photo of [your product], create a [square / vertical / wide] image for [Facebook feed / Google ad / website banner]. Keep the product's shape, color and markings unchanged. Setting: [plain light countertop, workshop bench, front step]. Colors: [your two brand colors]. Soft daylight, clean look, empty space on the [left / top] for text. Text in the image: "[exact words]".
Keep the wording in the image short. The model can render text, but Google lists text and fine detail among its known weak spots, so add long offers and phone numbers in an editor or in the ad's own text fields instead.
How do you get the right size for every placement?
Do not make one image and crop it three ways. Ask for each ratio in the prompt and export each one separately, then check that nothing important sits near an edge.
| Placement | Ratio to request | What to check |
|---|---|---|
| Facebook or Instagram feed post | Square, 1:1 | Subject centered, text clear of the edges |
| Stories or Reels | Vertical, 9:16 | Key content away from the top and bottom, where platform buttons sit |
| Google ad asset or website banner | Wide (wider than tall) | Subject still readable when the platform crops it |
Each platform changes its size rules from time to time, so confirm the current sizes in Google's image editor guidance and in your ad account before you export. Our guide to resizing and compressing images for your website covers file size for the web.
How do you catch mistakes before the image goes live?
Google's model documentation lists text, detail, factuality, localization and complex edits as areas where output can be wrong. Google Ads also says generated assets must be reviewed for accuracy, misleading claims and applicable law. Run this check on every image:
- Product accuracy: Compare it with the real item. Packaging, hardware, proportions and markings are where errors show up.
- Spelling: Read every word in the image, letter by letter.
- Faces and hands: Look for odd fingers, mismatched eyes or extra objects.
- Claims: The image must not show a result, a discount or a guarantee you do not offer.
- Match to your page: The image, the offer and the landing page should describe the same thing.
- Right crop: Open each export at the size where it will appear.
If an image fails, fix the cause. A wrong product shape usually means a better reference photo. Garbled words mean shorter text or adding the words in a design editor. A rejected asset means removing the likeness, logo or claim that triggered it.
What do Google Ads and Meta ask of you on AI labels?
Google Ads says AI-created or AI-edited assets may need disclosure under rules in the European Union, India and New York. You can add the label yourself or use Google's AI-label setting, and Google says that setting does not guarantee legal compliance. Images generated inside Google Ads carry machine-readable identification, including SynthID and C2PA metadata. Ordinary ad rules still apply, including rules on misleading content and inaccurate product representation, as set out in Google's ad policy pages.
Meta says its advertising and community rules apply whether or not AI made the content. In its AI labeling announcement, Meta describes labeling content when industry signals such as C2PA and IPTC, or self-disclosure, indicate AI use, and it says detection is not universal. Meta's ads transparency post adds that some AI-generated or significantly edited ad content, particularly photorealistic people, gets a label, and that political and social-issue ads carry extra disclosure requirements.
| Platform | What it says | What you do |
|---|---|---|
| Google Ads | AI assets may need disclosure in the EU, India and New York; generated assets are not guaranteed to pass Ads Policy | Use the AI-label setting or add a label, and review accuracy |
| Google-made images | Carry SynthID and C2PA identification | Leave the metadata intact when you upload |
| Meta (Facebook, Instagram) | Ad rules apply to AI and non-AI content alike; some AI content is labeled | Check the ad's AI disclosure controls before you publish |
No label replaces your own responsibility for advertising, consumer-protection, copyright, privacy and sector rules. If the ad touches a regulated field, check the specific rule before you run it.
What did Google announce on October 6, 2026?
Google's model card, published October 6, 2026, describes Nano Banana 2.1 as a high-efficiency image generation and conversational editing model based on Gemini 3.6 Flash. Google's developer documentation lists improved visual quality, prompt adherence, text rendering and subject consistency, output at 1K, 2K and 4K, up to 14 reference images, and web and image-search grounding.
Developers who call the model through the API pay per image, while the Gemini app and Google Ads tools run on your plan or your account's access. Google's pricing page lists these Standard API prices per image:
| Output size | Price per image |
|---|---|
| 1K | $0.0336 |
| 2K | $0.0504 |
| 4K | $0.113 |
A low per-image price is not the cost of a finished campaign. Reviewing, resizing, adding copy and getting approvals still take your time. Google does not say AI-generated images perform better than real photographs, so judge each image by whether it is accurate and clear.
When should you use a real photo instead?
Use real photographs whenever the customer needs to see the actual product, people, premises or finished work. A generated image can supply a clean background, a seasonal variation or a layout idea. It should not stand in for your crew, your storefront or a completed job. Our posts on stock photos or real photos and the rules for AI images in marketing cover that choice in more detail.
A hybrid often works best: photograph the real product once, then ask the model for new backgrounds and sizes around it.
For ad layouts, our guides to Facebook ad creative for a local business and to making a YouTube ad without filming show where generated images fit. If you also run a free AI assistant for captions and quotes, is free ChatGPT enough for your business helps you decide, and owners running Local Services Ads can read what to back up before LSA moves into Google Ads.
Is your AI ad image ready to run?
Pick an answer to begin.
1. You want a square Facebook post and a vertical Story from the same idea. What should you do?
2. Your generated image shows a misspelled word. What is the best fix?
3. Who is responsible for checking an AI image against ad rules and labels?
Frequently Asked Questions About nano banana for small business ads
Can I use Nano Banana 2.1 for business images?
Yes, within Google's terms, platform rules and applicable law. Review every image for accuracy before you use it.
Where can I use it without coding?
Google's pages point to the Gemini app, Google AI Studio and eligible Google Ads image tools.
How much does it cost?
In the API, Google lists $0.0336 per 1K image, $0.0504 per 2K image and $0.113 per 4K image at Standard pricing. In the Gemini app, access depends on your plan or account.
Should I upload a real product photo?
Yes, when the result must match the actual product. Use one product per reference image and several angles when you can.
Does Google Ads approve generated images automatically?
No. Google says generated assets are not guaranteed to pass Ads Policy, so check them for accuracy and misleading claims.
Do I need an AI label?
It depends on the region, platform, asset and campaign type. Use the labeling tools each platform provides and read the current rules before you publish.
The Bottom Line
Make one image this week: pick the job, upload a real photo, write the prompt with your brand colors and ratio, export each size, and run the review list. Save the prompt and the approved file together so the next batch starts from something that already worked.
You can do this yourself in Gemini, and for a basic post that is often enough. Keep real photos for people, your premises and finished work, and leave AI for backgrounds, variations and layouts.
If you want a repeatable setup with product accuracy checks, per-platform versions, label settings and approvals handled for you, I can build it. Web Leveling designs AI automation workflows for marketing assets, and you can reach me through the contact page to talk about your image workflow. We work with small and medium businesses across the country and overseas.
Terms
Terms
Tap a term to see what it means.
Nano Banana 2.1. Google's image generation and conversational editing model, described in its October 6, 2026 model card.
Reference image. A photo you upload so the model keeps your real product or style in the result.
Aspect ratio. The width-to-height shape of an image, such as 1:1 square or 9:16 vertical.
SynthID. Google's invisible watermarking system for AI-generated images.
C2PA. A provenance standard that attaches machine-readable information about how an image was made.
Asset Studio. The Google Ads area where eligible accounts can generate and edit campaign images.
