Karwa Chauth AI Photo Prompts for Gemini: Couple Edits

To get an accurate Karwa Chauth portrait from Gemini, upload one clear couple photo and write a prompt that names the exact outfit colours and fabrics, the ritual objects (puja thali, diya, a metal sieve, the moon) and what each person is doing, then tell it to keep both faces and skin tones exactly as they are. Gemini's photo editing feature (the one people call Nano Banana) handles the saree, sherwani, diya and moonlit backdrop well once it has that brief. Vague prompts like "make us look traditional for Karwa Chauth" produce generic Bollywood-poster results that your own family will spot as fake in two seconds. The fix is treating Gemini like a fussy photo editor you're briefing, not a magic wand.

Why Gemini and not a filter app

I get why people reach for filter apps first. They're fast and they're built for exactly one job. But a filter app applies the same overlay to everyone, which is why every Karwa Chauth Instagram post from the last three years looks like it came off the same production line: red saree, gold border, moon in the top corner, done.

Gemini works differently because you're writing instructions, not picking a preset. You can specify that the saree is a specific shade of maroon with a zari border rather than "red saree", that the husband's sherwani matches the wife's dupatta rather than clashing, that the thali has a diya, some rice, and a sieve (chalni) rather than a generic plate. That level of control is the whole point. Skip it and you're back to filter-app results, just with extra steps.

The actual prompts, written out in full

Copy these into Gemini alongside your uploaded photo. Swap the details in brackets for your own preferences.

• "Edit this couple photo to show them dressed for Karwa Chauth. She is wearing a deep maroon silk saree with a gold zari border and traditional jewellery including a maang tikka and bangles. He is wearing a matching mustard sherwani with a maroon dupatta or stole to coordinate. Keep both faces and skin tones exactly as in the original photo. Add a warm evening background with string lights and a rising moon visible in the upper corner."
• "Take this photo and add a decorated thali between them, held by her, containing a diya, a small mound of rice, and a metal sieve (chalni) resting beside it. Keep the pose natural, as if mid-conversation, not stiff or posed for a studio shot. Soft golden hour lighting, no harsh flash."
• "Generate a version of this couple where she is looking through a sieve at the moon and he is standing slightly behind her holding a diya. Background should be a home terrace or balcony at dusk, not a temple or studio backdrop. Keep the mehendi design on her hands consistent with the reference photo I uploaded separately."
• "Adjust the outfit colours to sage green saree with silver zari for her and a matching pale grey sherwani for him, everything else the same as the previous edit."

What makes those work is the detail. "Red saree" leaves the shade, fabric and border to Gemini, so it falls back on its red-and-gold default. "Deep maroon silk saree with a gold zari border" leaves it nothing to guess. The same goes for objects (sieve, thali, diya, never "traditional items") and for the line that keeps faces and skin tone unchanged, which is the one instruction you should never drop.

One copy-paste prompt with every blank filled

If you only run one edit, run this one. Nothing is left in brackets:

"Edit the uploaded couple photo into a Karwa Chauth evening portrait on a home terrace at dusk. Her: deep maroon silk saree, gold zari border, maang tikka, red bangles, mehendi on both hands. Him: mustard sherwani with a maroon stole. Props: a decorated puja thali with a lit diya, rice and a metal sieve. Keep both faces, skin tones and hairstyles exactly as in the photo. Natural, unretouched skin, soft warm ambient light, the moon low in the upper left."

The order matters. Identity lock first, outfit second, props third, light last. Put the face line first so nothing later in the prompt overrides it.

Build the scene from the real ritual, not a poster

Generic images show the objects without the moment. The day has a set order, and Wikipedia's Karva Chauth entry describes it: the fast runs from sunrise to moonrise, in Punjab the day starts with sargi, a pre-dawn meal traditionally given by the mother-in-law, and once the moon is up the wife looks at it, or its reflection in water, through a sieve with a small diya placed in its rim, then looks at her husband through the same sieve. Water is offered to the moon before the fast is broken. Each of those beats is its own scene:

• Sargi: "Pre-dawn kitchen, one warm lamp, she sits at the table with a plate of sargi her mother-in-law has set out, window still dark behind her. Keep her face exactly as in the photo."
• Moonrise: "She holds a metal sieve with a small lit diya in its rim up to her eyes and looks through it at the moon; he stands just behind her holding a steel glass of water."

"Sieve held up to her eye" beats "a sieve in the picture" every time. One is a pose; the other is a prop on a table.

A worked example

Say you run a small content page for an Indian wedding decor brand and you want a set of four Karwa Chauth couple images to post across the week leading up to the festival, without booking a photographer. You'd start with one clear, front-facing photo of a real couple (with their permission, ideally your own staff or a couple who've agreed to be featured), upload it to Gemini, and run the first prompt above. You get an image back quickly. It's close but the sherwani colour is off, so you refine with a second instruction: "keep everything the same but change his sherwani to a deep bottle green." That single change is one more generation, and because it names only the sherwani, the rest stays put.

Then you run the thali version and the sieve-and-moon version separately, so you end up with four distinct images from one source photo rather than four photos that are obviously the same edit with a filter swapped. Crop each one as a portrait for the feed and as a full-height 9:16 frame for Stories. That's a full week of content from one photo and one short session of prompt refining, which is the real pitch here, not the aesthetics.

The part most guides skip

Every one of these prompts produces something Bollywood poster shaped: symmetrical face, glossy skin, jewellery that looks studio-lit even when the background says "home terrace." That's not a bug in your prompt, it's the polished look these image models fall back on unless you tell them otherwise. When all you want is a different setting behind the couple, background change prompts for Gemini do that job. If your family's actual Karwa Chauth photos have always been a bit blurry, a bit overexposed from the kitchen light, a bit real, then the AI version will look conspicuously polished next to them. Some people love that. Others post it and get a comment from an aunt asking why they look "different." The fix is one line: add "natural, unretouched skin, soft ambient light, no studio gloss" to the prompt, then put the result next to a real photo from last year's Karwa Chauth and check the skin, the light and the jewellery look like the same family.

There's a second thing worth saying plainly: Karwa Chauth isn't one uniform aesthetic. Customs vary by region and by family, from what's eaten at sargi to the outfits and the items on the thali. Gemini, left to its own defaults, will give you the most familiar look it knows, usually the red-and-gold, Bollywood-styled version. If your own family's traditions look different, you need to say so explicitly in the prompt, right down to naming the region or the specific ritual object. The tool won't know unless you tell it, and it will happily hand you a confident, well-lit, slightly wrong answer otherwise.

What goes wrong, and the one-line fix

• The sieve turns into a plate, a mirror or a flower. Describe it: "a metal chalni, a flat round sieve with a fine mesh, held up to her eye."
• Hands go wrong or the mehendi disappears. Add "both hands visible, mehendi design unchanged from the reference photo," and upload a clear photo of the hands as a second image.
• Faces drift lighter, slimmer or smoother. Re-run with "keep both faces and skin tones identical to the original, no beautifying, no skin smoothing."
• You asked for three changes and got a muddle. Go back a step and ask for one change per follow-up, naming the single object you want changed.
• Jewellery melts into the neck or ears. Ask for "jewellery sitting on top of the skin, with visible gaps and soft shadows."

If you're wondering what to write in Gemini for photo editing, I've shared prompts you can copy.

Privacy, before you upload anything

This is the bit that gets left out of most lists like this. When you upload a personal photo of yourself or your partner to any AI photo tool, including Gemini, you're sending that image to a company's servers. Check the current settings in your Google account for whether your uploads can be used to improve the model, and decide if you're comfortable with that for a photo of your face, your partner's face, and possibly your home in the background. For a decor brand's stock content shoot this is a non-issue. For a private couple's photo you're planning to send round to family, it's worth the thirty seconds it takes to check the setting.

A quick check before you upload:

• Open Gemini's settings and look for the activity control, the setting that decides whether your chats are kept. The label and its place in the menu change between app versions, so search the settings for "activity" if you can't see it.
• Turn it off, or set it to auto-delete, if you don't want the photo kept with your history.
• Crop out anything you'd rather not share before you upload: a house number, a car plate, children in the background.

Step by step, start to finish

• Choose one clear, well-lit source photo, front-facing, both faces visible, nothing else distracting in frame.
• Open Gemini and upload that photo along with your written prompt, keeping colour, object, and pose instructions specific rather than general.
• Generate the first version and check three things: are both faces still recognisable, does the outfit colour match what you asked for, does the pose look natural rather than stiff.
• If something's off, don't start over, refine with a follow-up instruction referencing only the change you want, such as "keep everything the same but soften the lighting."
• Once you're happy, save it and crop it for where it's going: a portrait crop for a feed post, a full-height 9:16 frame for Stories. For a Reels cover, Instagram's reel size page recommends 420 x 654 pixels, a 1:1.55 ratio.
• Run a second and third variation from the same source photo for different outfit colours or props, so your content doesn't look like one image reused four times.

Solo and group versions

Not every Karwa Chauth photo is a couple, and with one face in the frame the identity line matters even more:

• Solo: "Edit this photo of one woman into a Karwa Chauth portrait on a balcony at golden hour. Red Banarasi saree with gold border, gold jhumkas, sindoor, mehendi on both hands. She holds a metal sieve with a lit diya in its rim. Keep her face, skin tone and hair exactly as in the photo, natural unretouched skin."
• Friends: "Five women sit in a circle on a rug, each in a different jewel-toned saree, each with her own decorated puja thali and a lit diya. Keep every face exactly as in the uploaded group photo."

I keep a separate page on Gemini AI photo prompts for girls, so you don't have to dig.

What to avoid

• Prompts that say "traditional Indian outfit" with no colour or fabric specified. You'll get whatever the model's default is, which is usually red and gold regardless of what you wanted.
• Skipping the instruction to preserve the original face and skin tone. Without it, some edits drift toward a smoothed, lighter, more generic face that doesn't look like either person anymore.
• Posting the result without telling anyone it's AI-edited, if it's going somewhere public. Not a legal issue in most cases, just an honesty one, and it saves you an awkward comment thread later.
• Using a photo of a couple who haven't agreed to have their image edited and shared, even if the intent is a nice gesture.

I keep every related walkthrough in the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps. To write your own, try the free ChatGPT prompt generator.

Frequently asked questions

Is Gemini free to use for these Karwa Chauth photo edits?

Gemini has a free tier that supports photo uploads and image edits. Google doesn't publish one fixed daily figure for image edits: the cap depends on your plan and country and changes over time, and the app tells you when you've reached it. For a handful of Karwa Chauth variations from one photo, the free tier is usually enough.

What time does the fast end, and when should the moon photo be set?

The fast runs from sunrise to moonrise and is broken once the moon has been seen. Moonrise differs by city and by year, so check a local moonrise time for your town before you stage a real moon shot.

Can I use a photo of just one person and add a partner who isn't in it?

You can ask Gemini to generate a second figure standing beside the uploaded photo, but the result tends to look noticeably artificial since the model is guessing an entire person's appearance, pose, and lighting to match. It works better as a stylised graphic than as something meant to pass as a real photo.

Can Gemini keep my mehendi design?

Only if you give it something to keep. Upload a clear, close photo of both hands alongside the main photo and tell it "mehendi design unchanged from the reference." Without that it redraws the pattern, so zoom in on both hands after every edit.

If you're curious about how Perchance creates AI images, I've covered that too.

Where to check the details

For Gemini's own limits and privacy controls, check the settings inside the Gemini app; the two links below are general reading on AI, not guides to Gemini's photo tool.

• Anthropic's own documentation
• the NIST AI Risk Management Framework