Gemini Image Generation and Editing with AI Pro

The model doing the work

Image generation and editing in the Gemini app run on Google's image model, currently called Nano Banana 2. You describe an image, or upload one and describe the change you want, and the model returns a picture you can keep iterating on in the same conversation. The same family of models powers image work inside Google Flow, which is how a still frame becomes the starting point for a video.

It is available in every country where the Gemini app is available, Nepal included, and it is restricted to users aged 18 and over.

What a subscription changes

This is the part worth being precise about, because image generation is not a paid-only feature. Free accounts can already make images. A Google AI plan adds a better model to the list you can pick from.

Model in the Gemini app	Free account	Google AI Pro
Nano Banana 2 Fast	Yes	Yes
Nano Banana 2 Thinking	Yes	Yes
Nano Banana Pro, also called Gemini 3 Pro Image	No	Yes
Output resolution on the Pro model	n/a	1K, 2K and 4K
Higher limits on Google Photos generative features	No	Yes

Google publishes no numbers for how many images you can generate on either tier. It says limits apply and that they can change, which is the honest answer anyone selling you the plan should give. Subscribers to AI Plus and AI Ultra also get the Pro image model; the difference between those tiers sits elsewhere, as set out in Plus vs Pro vs Ultra.

The two things people notice when they move to the Pro model are the higher output resolutions, which matter if the image is going to print, and noticeably better text rendering inside the image.

Editing a photo you already have

Generation gets the attention, but editing is where most small businesses get value. You are not starting from nothing, you are fixing a photograph you already took.

Upload the original. A product shot, a shopfront, a portrait, a screenshot of a design.
Describe the change in plain terms. "Remove the background and place this on a plain warm grey studio backdrop, keep the product exactly as it is." Naming what must not change matters as much as naming what should.
Iterate in the same thread. Ask for the next adjustment rather than starting a new prompt, so the model keeps the context of what you are working on.
Check the details that break. Text on packaging, logos, hands, jewellery and fine patterns are where an edit quietly ruins a photo. Zoom in before you publish.
Export at the resolution you need. The Pro model's 2K and 4K output is the reason to use it for anything larger than a social post.

Text inside an image

Better text rendering is the headline improvement in the Pro image model, and it is the reason posters, offer cards and thumbnail concepts became practical. It is also where the caution belongs.

Do not assume Devanagari works. Google makes no claim about Nepali text rendering in generated images. Test it with your own words before you build a workflow on it, and be ready to generate the artwork without text and add Nepali headlines in Canva, Photoshop or any editor that handles the font properly.

English headlines on a poster generally hold together well enough to use, but proofread every character. The model draws letters, it does not typeset them.

What people here are actually making

Festival and offer posters

A shop needs a Tihar discount graphic on Friday and again on Saturday. Generated backgrounds and layouts get a draft on screen in minutes, with the final Nepali text set in a design tool.

Product images for listings

Clean backdrops, consistent lighting across a range of items, lifestyle variants of the same product. Useful for an online store or a Daraz listing where a real studio shoot is not worth it.

Thumbnail concepts

Creators use it to test five thumbnail ideas before committing, then rebuild the winner with their own photograph. Our YouTuber workflow covers where this sits in a week of production.

Mock-ups and pitches

Showing a client what a signboard, a menu redesign or a packaging idea could look like, before anyone commissions the real design work.

Storyboard frames

Generate the frames of a scene, then feed the good ones into Flow as ingredients or starting frames so the video matches the look you approved.

Social graphics for services

A trekking operator or a consultancy that posts daily needs images that are not stock photos everyone else is also using.

Watermarks, honesty and the limits of a generated picture

Google marks images produced in the Gemini app with SynthID, its invisible digital watermark, along with a visible watermark on the output. Assume any image you generate can be identified as AI-made, and plan your communication around that rather than against it.

Two practical rules. Do not present a generated image as a photograph of a real place, room, dish or person your customer will later encounter, because a hotel or restaurant listing that does this creates a complaint, not a booking. And do not generate images of real people, staff or customers without their agreement.

The other limit is consistency. Asking for the same character or the same product across ten images is harder than getting one good image, which is exactly the problem video generation runs into as well. Supply a reference image every time rather than trusting the model to remember.

From a still image to a moving one

Two Google features connect images to video, and both improve on a paid plan. Inside Flow, an image can be the starting frame for a generated clip or an ingredient that keeps a character consistent across shots. Inside Google Photos, the generative features Photo to video and Remix have higher limits for AI Pro subscribers, which is the low-effort version of the same idea applied to your own photographs.

If image work and short video are both part of what you do each week, the same subscription covers both, and our guide for content creators and agencies shows how the pieces fit into a delivery schedule. WebsNP sells that plan for your own Gmail address, paid in rupees, if paying Google in dollars is the part that has been stopping you.

Frequently asked questions

Can free Gemini accounts generate images?

Yes. Free accounts get the Fast and Thinking versions of the image model. A Google AI plan adds the Pro image model, Nano Banana Pro, with 1K, 2K and 4K output and stronger text rendering. Limits apply on every tier and Google does not publish the numbers.

How many images can I generate per day with AI Pro?

Google does not publish a figure. Its help pages state that limits apply to image generation and that limits can change over time. Treat any specific daily number you see quoted elsewhere as unverified.

Does it write Nepali text on images?

Google makes no claim about Nepali or Devanagari rendering, so test it yourself before relying on it. The safer method is to generate the artwork without text and add Nepali headlines in a design tool where you control the font.

Are generated images watermarked?

Yes. Google applies SynthID, an invisible digital watermark, and a visible watermark to images from the Gemini app. Plan on your image being identifiable as AI-generated.

Can I use generated images for my business advertising?

They are commonly used for concepts, backgrounds, posters and mock-ups. Do not use a generated image to depict a real room, dish, product or person your customer will actually encounter, and check the current Gemini Apps terms for the usage rules that apply to your account.

Is image generation available in Nepal?

Yes. Google states image generation is available in every country where the Gemini app is available, and Nepal is on that list. You must be 18 or older.

Sources and further reading

Product facts on this page were checked against Google's own documentation on the date shown above. Google changes plan contents, limits and country availability over time, so treat the official pages as the final word.