How AI Photo Editing Works: Profile Photo, Festival Look and 3D Figurine in Gemini (Nano Banana)

By Ravindra Bagale, trainer · LinkedIn · Last updated: 19 October 2026

Generative AI course · Chapter 23 · Estimated reading time: 11 minutes

1. Suppose we need a profile photo by tomorrow

Suppose we are Neha, an HR executive in Pune. We are updating our LinkedIn profile tonight. The only good photo of us is a phone selfie in our bedroom: a cupboard behind us, yellow tube light, a T-shirt.

We open Gemini, upload the selfie and type:

Make this a professional photo

The result looks like a magazine cover. The background is an office, we wear a black blazer, our skin is smooth like plastic, and our face is about 80% ours. A recruiter who meets us on a video call will notice.

The better way is to tell the AI what must stay the same, change one thing at a time, and zoom in before we accept a result.

2. How AI photo editing works

Old photo apps change the pixels you select: crop, brightness, a filter. An AI photo editor works differently:

1. It looks at your photo and reads your instruction.
2. It draws a new image that follows both.
3. Parts you did not mention are copied closely, but not perfectly, because they are drawn again too.

That is why small details drift: earrings, teeth, a mole, the shape of your eyes. Every edit is a new drawing. The more you ask for in one go, the more the model has to re-imagine, and the more your face can change.

There are two kinds of jobs:

Editing a profile photo is an edit job. A 3D figurine is a create job. Expect a lower likeness in the second.

3. Which Gemini image model you get (October 2026)

"Nano Banana" is the nickname of Google's Gemini image model. On Google's Gemini image page (checked 19 October 2026):

• Free accounts get Nano Banana 2.
• Google AI plans unlock Nano Banana 2.1, which Google describes as its most advanced image model, with better quality and more consistent results.

To open it:

1. Open the Gemini app or gemini.google.com and sign in with your personal Google account (image editing needs you to be 18 or older).
2. Tap the tools button near the message box and choose Create images.
3. Tap + to upload your photo.
4. Type your instruction and send it.

Gemini can also edit using more than one uploaded image, for example your photo and a photo of an outfit. We use this in worked example 2.

4. Worked example 1: a profile photo that still looks like you

Prompt 1 (background only):

This is my photo. Keep my face, hair, expression, skin texture and earrings
exactly the same. Change only the background to a plain light grey wall.
Do not smooth my skin. Do not change my clothes yet.

Expected output (yours will be similar, not the same): the same selfie with the cupboard gone and a soft grey wall behind. The tube light colour is still yellow on the face.

Prompt 2 (light), in the same chat:

Keep everything from the last image. Change only the light: soft daylight from
a window on the left, natural skin tones. Keep my face exactly the same.

Prompt 3 (clothes), in the same chat:

Keep everything from the last image. Replace the T-shirt with a plain navy
blue blazer over a white top. Head-and-shoulders crop, square 1:1.

Expected output (yours will be similar, not the same): a clean head-and-shoulders photo with grey background, window light and a navy blazer, with the face close to the original.

Now do the zoom check (section 7). If the face changed, go back one step: "Use the image from 2 steps ago and only change the T-shirt to a navy blazer."

One honest rule: a profile photo should still look like you on a video call. Fixing the background and light is fine. Changing your face shape, skin colour or age is not, and it can break trust.

5. Worked example 2: a festival look with your own outfit

Suppose Neha wants a Diwali greeting photo wearing her own green silk saree, which is folded in the cupboard. She takes 2 photos: one of herself, standing, and one of the saree spread on the bed.

Steps:

1. In Create images, tap + and upload both photos.
2. Say which photo is which, and what to take from each.

Prompt:

Image 1 is me. Image 2 is my green silk saree with a gold border.
Show me wearing the saree from image 2, draped in a simple Maharashtrian style,
standing on a balcony with diyas and soft evening light.
Keep my face, hairstyle and expression exactly as in image 1.
Keep the saree's colour, border and pattern exactly as in image 2.
Vertical 4:5 for Instagram.

Expected output (yours will be similar, not the same): Neha on a balcony at dusk with diyas, wearing a green saree with a gold border. Zoom in on the border pattern: the model often simplifies a complex pattern. If it changed, say: "Keep everything, but copy the border pattern from image 2 more exactly."

Using your own outfit photo makes the result honest and personal, and the model has a clear picture to follow instead of inventing a saree.

6. Worked example 3: a 3D figurine (a create job)

The "3D figurine" trend turns your photo into a toy version of you on a desk. This is a create from reference job, so the likeness will be lower than in an edit.

Prompt:

Create a 3D collectible figurine of the person in this photo, standing on a
round clear acrylic base on a wooden office desk. Next to it, a toy box with
the figurine's picture on it. Keep her hairstyle, glasses and navy blazer.
Realistic product photo lighting. Square 1:1. No text on the box.

Expected output (yours will be similar, not the same): a glossy toy figure with Neha's hairstyle, glasses and blazer, on a clear base, with a box beside it.

Why "No text on the box"? AI images often get text wrong (missing letters, extra letters). If you want a name on the box, add it later in a photo editor where you control the spelling.

7. The zoom check before you post

Before you accept any AI photo, open it at full size and check, one by one:

1. Eyes: same shape and colour, looking the same way.
2. Teeth and smile: not too white, not a different smile.
3. Ears, earrings, glasses: still there and the same style.
4. Hands and fingers: 5 fingers each, no melted shapes.
5. Skin: natural texture, not plastic.
6. Edges: hair and shoulders blend into the background without a halo.
7. Text: any word in the image spelt correctly.

If 2 or more fail, go back one step in the chat or start again from your original photo.

8. Other tools that edit photos with words

The same ideas (say what stays, one change at a time, zoom check) work in other tools:

• ChatGPT: upload a photo and ask for an edit; your images are saved under My images in ChatGPT.
• Google Photos: has its own AI editing features on supported phones.

Image models and their names change often, so check the tool's help page if a button is missing.

9. Consent, labels and where AI photos must never be used

• Only edit photos of people who agreed, and never a stranger's photo. Fake images of real people can hurt them and can be a crime.
• Gemini images carry a visible watermark and an invisible SynthID watermark. Do not try to remove them.
• Label realistic AI photos. India's IT Rules amendment of 2026 asks platforms to clearly label synthetically generated images. Say "AI-edited" in your caption when the photo looks real.
• Never use an AI-edited photo for documents like Aadhaar, PAN, passport, visa, exam forms or an office ID card. They need a real, unedited photo.
• Do not upload children's photos or ID cards to any AI tool.

10. Beginner mistakes and quick fixes

11. Try at home

Recap

Edit one photo tonight with the one-change-per-message method and do the zoom check. You will see exactly where the AI drifts, and how to stop it.

Previous chapter: AI for Small Business Accounts: GST Notice Replies, Invoice Checks and Sales Reports (with a CA Check)

Next chapter: ChatGPT Image Generation: Posters, Ghibli-Style Art and Thumbnails

Ravindra Bagale, trainer: linkedin.com/in/ravindra-bagale. The person and photos described on this page are made up for learning. Gemini image model names and features were checked on Google's Gemini image page and Gemini help pages on 19 October 2026 and can change. AI images change every time, so your output will be similar, not the same.