← AI prompts · Trending · Diwali
Diwali saree AI photo prompt — lamps behind, soft light
The most searched of the Diwali looks. This version puts the lamps behind you so the flames fall into bokeh and your face gets a softer fill. Copy this free ChatGPT & Gemini prompt, upload one photo, and it keeps your own face.
Use my uploaded portrait as the primary and exact facial-identity reference. Create a vertical 9:16 photorealistic Diwali portrait of me standing in a softly lit Indian home at night, wearing a saree.
Preserve my face exactly as it appears in the reference photograph. 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 or reinterpret my face. The result must remain unmistakably the same person.
Place the lit clay oil lamps BEHIND me rather than in front, arranged along a ledge, stair edge or windowsill at roughly shoulder height and further back. Let them fall well out of focus so each flame becomes a soft warm circle of bokeh. Light my face with a gentle, diffuse warm fill from the front left, low in intensity, so the portrait is softly lit rather than dramatically lit.
Dress me in a silk saree with a visible weave and a genuine drape: pleats falling cleanly at the front, the pallu carried over the left shoulder and hanging in soft folds down the back, and a fitted blouse with a neat sleeve. Choose a deep jewel tone such as emerald, wine, indigo or mustard with a contrasting border. The fabric must behave like real cloth, with weight, creasing and a visible edge where it folds. Do not render it as a flat printed costume.
Keep jewellery restrained and realistic: small jhumkas, a thin chain, a few bangles. Avoid heavy, ornate or fantasy jewellery.
Smartphone portrait photography, warm Diwali atmosphere, authentic Indian festive aesthetic, realistic skin texture, realistic silk texture, high facial detail, shallow depth of field with strong background bokeh, 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, 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, costume-like clothing, ill-fitting garment, dropped shoulder seam, floating flame, doubled flame, malformed diya, incorrect anatomy, extra fingers, visible seam, 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 the saree version of the Diwali portrait, which is the most searched of the set. You upload one clear photo and the model keeps your real face while building the saree, the warm room and the lamps behind you.
Two things in this prompt are deliberate. The lamps sit behind you, not in front — the flames become background bokeh and your face gets a soft, forgiving fill rather than hard light from below. It is far easier to get a usable result this way than with the close diya portrait.
And the drape is named, not left open. Asked for “a saree”, these models reliably return something that reads as a costume: flat fabric with a printed-on look and no real folds. Naming the pleats, the pallu over the left shoulder and the weave is what makes it behave like cloth.
It is tuned for identity-locked, image-to-image tools: ChatGPT (image mode), Google Gemini, Nano Banana and similar, at 9:16.
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 saree looks printed on. Flat fabric with no weight, no folds and an edge that does not behave like cloth. This is the commonest failure and the reason the prompt names the weave, the pleats and the pallu. Re-run rather than accept it.
- The jewellery goes ornate. Left to itself the model tends towards heavy, fantasy jewellery that pulls the whole image towards costume. The prompt asks for small jhumkas and a thin chain; if it ignores that, say so again at the end.
- The lamps drift in front. If the model puts the diyas back in front of you, the soft fill is gone and you get the hard under-lighting of the close portrait instead. Re-run, and if it persists add all lamps are behind the subject and out of focus.
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.
Want the hard-lit version instead? The diya portrait puts the lamps in front for a dramatic close-up. The rangoli doorway is the full-length frame, and the couple photo handles two people. All six are in the Diwali AI photo editing prompts roundup.
Frequently asked questions.
What is the Diwali saree AI photo prompt?
It is a copy-paste AI image prompt for a vertical 9:16 Diwali portrait in a saree, with the oil lamps placed behind you so the flames fall out of focus and your face gets a soft warm fill. You upload one photo and the prompt keeps your real face while building the saree, the jewellery and the room.
Why does my saree come out looking like a costume?
Because asked for just a saree, these models return flat fabric with a printed-on look and no real folds. This prompt names the weave, the pleats at the front and the pallu carried over the left shoulder specifically to stop that, since describing how cloth behaves is what makes the model render it as cloth rather than as a pattern.
Why are the lamps behind instead of in front?
It is a much more forgiving light. Lamps in front light the face from below, which is dramatic but difficult, and it is the setup most likely to drift off your features. Behind you, the flames become soft bokeh and your face takes a gentle front fill instead, which is far easier to get a usable result from.
Do I need to own a saree to use this?
No. The saree, the jewellery, the lamps and the room are all built by the prompt around the face in your uploaded photograph. An ordinary indoor photo of yourself is enough to start with, and a sharp front-facing one gives the best result.
Is the Diwali saree 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 yourself. 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.