How to Write AI Photo Prompts That Work

The eight parts of a photo prompt

Google's prompt tips for Nano Banana Pro list subject, composition, action, location and style as the building blocks, and suggest adding camera, lighting and the canvas shape. For editing your own photo, two more parts matter a lot: what to keep unchanged, and what to avoid. Put together, that gives eight parts.

Parts 1 to 3: subject, outfit and scene

Subject. When you edit your own photo, start with "Using the uploaded photo" so the model knows to work from your picture rather than invent a new person. Then say who: "the man from the photo", "the couple from the photo". For groups, say how many people should appear so nobody is added or dropped.

Outfit. Specific beats general. "Kurta" is fine, but "an off-white cotton kurta with fine chikankari embroidery and a maroon stole" gives the model fabric, colour and detail to follow. For regional wear, name it properly: Kanjeevaram, Paithani, Phulkari, Bandhani, Nehru jacket. Our traditional wear prompts show many examples.

Scene. Pick one clear place and two or three supporting details. "A Lohri bonfire at night with people clapping, softly blurred behind" is better than a long list of ten objects, which tends to produce clutter. If the scene is not important, choose a plain studio backdrop so attention stays on you.

Parts 4 to 6: lighting, style and framing

Lighting. Light sets the mood more than any other word. Golden-hour sunlight gives warm, soft skin. Soft window light gives a calm, natural portrait. Low-key light gives drama with deep shadows. Google's prompt tips give an example like "golden hour backlighting creating long shadows", which names both the time and the effect. Our glossary explains terms like Rembrandt lighting and rim light.

Style. Say what kind of image it is: an ultra-realistic photo, a cinematic film still, a vintage film photo, a black and white portrait, a 3D animated character or an oil painting. Then add one or two finishing details that fit the style, like "slight film grain and faded colours" for a retro Bollywood look.

Framing. Tell the model how much of you to show. Close-up from the shoulders up suits a profile picture. Half-body suits outfits. Full-body suits sarees, lehengas and poses with a car or bike. Wide framing shows the setting. Camera words such as "85mm portrait lens look" or "eye-level angle" help push the result towards a natural photograph.

Parts 7 and 8: constraints and ratio

Constraints are the guard rails. For your own photo, the most important one is the keep-face line: "Keep the face exactly as in the uploaded photo: same features, skin tone, hairstyle and age. Do not beautify or slim." Then add what to avoid: "Correct hands and fingers, no text, no watermark, no extra people." Read our guide on keeping your face the same for stronger versions.

Ratio is the shape. Put it at the end of the prompt so it is not lost: 4:5 for an Instagram feed post, 9:16 for stories, reels and WhatsApp status, 1:1 for a profile picture. OpenAI's help page says ChatGPT also has an aspect ratio picker. See our aspect ratio guide for the full list.

A weak prompt rewritten, step by step

Here is a typical first attempt and how each part improves it.

• Weak: "Make a nice Christmas photo of me."
• Add subject and outfit: "Using the uploaded photo, create a portrait of the man from the photo wearing a dark green knitted sweater and a grey scarf."
• Add scene and light: "...sitting beside a decorated Christmas tree with warm fairy lights, in soft evening lamp light."
• Add style and framing: "...as a realistic photo with natural skin texture, half-body framing at eye level."
• Add constraints and ratio: "...Keep my face, beard and skin tone exactly as in the photo. Correct hands, no text, no extra people. Vertical 4:5."

The final version is longer, but every phrase gives the model a decision it would otherwise make on its own. For more ready examples, see our Christmas prompts and prompts for boys.

Habits that make prompts better

• Write in plain sentences. Google's Nano Banana tips say simple, conversational instructions work, so you do not need strings of keywords.
• One idea per sentence. It is easier for you to edit and for the model to follow.
• Cut contradictions. "Bright daylight" and "moody night" in one prompt will confuse the result.
• Avoid praise words. "Stunning", "flawless" and "perfect" invite beautifying rather than accuracy.
• Save prompts that worked for you. Keep them in your notes and change only the outfit or scene next time.
• Iterate in follow-ups. After the first image, ask for one change at a time rather than pasting a new long prompt.