← AI prompts · Trending · Diwali
Diwali couple portrait AI photo prompt — keeps both faces
The Diwali couple portrait — two people in festive clothes, warm lamp light, a 4:5 frame built for the grid. Copy this free ChatGPT & Gemini prompt, upload one photo of the two of you, and it keeps both faces.
Use my uploaded photograph as the primary and exact facial-identity reference for BOTH people in it. Create a vertical 4:5 photorealistic Diwali couple portrait.
Preserve both faces exactly as they appear in the reference photograph. For each person independently, retain the original facial structure, eye shape, eyebrows, nose, lips, smile, teeth, jawline, cheeks, skin tone, skin texture, pores, facial hair, hairline, age, natural asymmetry, marks and other identifying details. Do not redesign, beautify, smooth, reshape, swap, merge or average the two faces. Do not make them resemble each other. Both people must remain unmistakably themselves.
Keep the two people at their relative heights and build from the reference photograph. Position them standing close together and turned slightly towards each other, shoulders overlapping, both faces clearly visible and similarly lit.
Dress them in coordinated traditional Indian festive clothing that is not identical: deep jewel tones that sit well together, realistic silk or cotton with visible weave and natural creasing, understated jewellery. Match the formality of the two outfits to each other.
Light the scene with warm practical light as if from diyas and string lights just out of frame, soft and even across both faces, with a gentle amber falloff. Behind them, a softly out-of-focus Indian home at night: a doorway, marigold strings, small points of lamp light rendered as soft bokeh.
Smartphone portrait photography, warm Diwali atmosphere, authentic Indian festive aesthetic, realistic skin texture, high facial detail on both subjects, shallow depth of field, natural contrast, high-resolution photographic result.
Do not generate lettering or a watermark.
NEGATIVE PROMPT
altered identity, different person, face swap artifacts, changed facial proportions, beautified face, plastic skin, excessive skin smoothing, changed eye shape, changed nose, changed lips, changed jawline, artificial smile, incorrect teeth, extra teeth, distorted mouth, asymmetrical eyes, duplicated features, de-aged face, aged face, different hairstyle, changed skin tone, cartoon face, painting, illustration, 3D render, waxy texture, blurred face, low-detail face, oversharpening, excessive HDR, blown-out highlights, melted candle wax, malformed diya, floating flame, doubled flame, incorrect anatomy, extra fingers, visible seam, collage border, illegible generated text, logo, watermark
SETTINGS FOR MAXIMUM FACE PRESERVATION
Use image-to-image or reference-image editing, not text-only generation.
- Keep denoise or transformation strength around 0.15-0.30.
- Set identity/reference strength high, around 80-95%, where supported.
- Mask or protect the entire face, hairline and visible ear.
- Generate or replace only the clothing, lighting and surroundings.
- Disable automatic face enhancement, beauty filters and skin smoothing.
- Add any caption separately after generation for clean, accurate typography.
About this prompt
This is a face-reference, identity-preserving AI prompt for a two-person Diwali portrait. You upload one photograph containing both people, and the prompt keeps both faces while rebuilding the clothing, the warm lamp light and the home behind you.
Two identities in one frame is genuinely harder than one, and the prompt is written around that single problem. It instructs the model to preserve each face independently, and explicitly forbids the two most common failures: averaging the faces towards each other, and swapping features between them. Most of the extra wording in this prompt exists for that reason.
It is tuned for identity-locked, image-to-image tools: ChatGPT (image mode), Google Gemini, Nano Banana and similar. It is built as a 4:5 vertical, which is the largest frame the Instagram grid will show without cropping.
How to use it
- Tap Copy prompt above.
- Open ChatGPT (image generation), Google Gemini or Nano Banana in image-to-image / reference mode, paste the prompt, and attach one clear, front-facing, well-lit photograph.
- Generate at the ratio named in the prompt.
- Re-run once or twice — the first result is rarely the best one.
The three ways this goes wrong
- The faces drift towards each other. Image models have a habit of averaging two people in one frame until they look like siblings. The prompt forbids it in as many words, but if it still happens, upload a photograph where the two faces are further apart and more clearly separated.
- One face is sharp and the other is not. Usually a problem in the source: if one person is further back, turned away or in shadow, the model has less to hold on to. Pick a reference where both faces are similarly lit and similarly sized.
- The outfits come out matching. Coordinated reads well; identical reads like a costume. If the model puts them in the same colour, re-run or name the two colours yourself at the end of the prompt.
Tips for the best result
- Say the face lock rather than set it. ChatGPT, Gemini and Nano Banana have no transformation-strength or identity-strength control — the prompt's settings block is written for image-to-image apps that do, where a denoise or strength slider around 0.15–0.30 is a sensible start. Everywhere else the control is the sentence: keep my exact face, build only the scene around me, do not restyle my features.
- There is no beautify switch to turn off in those three tools either. The one that matters sits on the phone camera or gallery app that took your reference photo — upload the unretouched original, because AI image tools have a well-known habit of building on whatever smoothing is already in the file.
- Use a sharp, front-lit photo so the face lock has clear features to hold.
- Add any caption afterwards in your editor rather than asking the model for text, which is why the prompt forbids generated lettering.
On your own? The diya portrait is the warmest single-person frame, and the rangoli doorway shows the whole outfit. All six looks are in the Diwali AI photo editing prompts roundup, or browse the full AI photo prompts library.
Frequently asked questions.
What is the Diwali couple portrait AI prompt?
It is a copy-paste AI image prompt for a vertical 4:5 Diwali portrait of two people in festive clothing, lit warmly as if by diyas and string lights just out of frame. You upload one photograph containing both people and the prompt keeps both real faces while building the clothing, lighting and the home behind them.
Does it really keep both faces?
That is what most of the prompt is for. It instructs the model to preserve each face independently and explicitly forbids averaging the two together, swapping features between them or making them resemble each other. Image tools have a known habit of blending two people in one frame, so a photograph where both faces are clearly separated and similarly lit will give the best result.
What photo should I upload?
One photograph containing both people, front-facing, sharp and evenly lit, with both faces a similar size in the frame. Avoid pictures where one person is much further back, turned away or in shadow, because the model then has far less to hold on to for that face and only one of the two will come out looking right.
What aspect ratio does this use?
It is written for 4:5 vertical rather than 9:16, because two people side by side need the width, and 4:5 is the tallest frame the Instagram grid displays without cropping. You can ask for 9:16 instead if you want it for a Story, though the two faces will sit smaller in the frame.
Is the Diwali couple portrait prompt free?
Yes, the prompt is free to copy and use. You only need an AI image tool such as ChatGPT, Google Gemini or Nano Banana and one clear photo of the two of you. There is nothing to buy.
Keep going
More trending AI photo prompts.
Copy-paste prompts blowing up on Instagram — all free, all keep your face.