How to write AI photo prompts: a simple structure to follow

A photo editing prompt starts from your own picture, so it must say what stays and what changes. Six short parts are enough for most portraits.

• Identity: your face, hair and skin tone stay as they are in your photo
• Light: where it comes from and how soft it is
• Wardrobe: what you wear, with fabric and color
• Setting: where you are, in a few plain words
• Framing: how much of you is in the picture and how you sit or stand
• Finish: skin texture, grain and overall color

Start with identity, because it protects the thing you care about most. Put light second, since it sets the mood and shapes the face. Then add wardrobe, setting, framing and finish. Treat this order as a sensible default, not a rule, because tools read prompts differently.

Give each part one short phrase. Six concrete phrases are easier for the tool to follow than a paragraph of adjectives, which hides the instructions that matter. If you must cut something, cut the finish first and keep the identity line.

Concrete words describe something a camera could capture: soft window light from the left, a gray crew-neck sweater, a pale wall. Vague words such as beautiful, high quality or masterpiece describe nothing in particular, and tags such as 8K may add little.

Avoid wording that edits the face, such as flawless, airbrushed or younger. Avoid stacking genre names like cinematic, editorial and glamorous in one line, because they compete. Avoid naming real people as a reference, which can pull the face toward theirs. Lead with what you want, and add a short exclusion, such as no text or logos, only when something keeps appearing.

Here is an original example for a window-lit portrait in a gray sweater. It is generic and only shows the structure, so change every detail to match your own photo.

Join the six lines below into one paragraph when you paste them into your tool. The labels are for you, not for the tool.

• Identity: Keep the face, hair and skin tone from my attached photo unchanged.
• Light: Soft daylight from a large window on the left, with gentle shadow on the right side of the face.
• Wardrobe: A plain gray crew-neck sweater.
• Setting: A plain room with a pale wall.
• Framing: Chest-up, shoulders turned slightly, calm expression.
• Finish: Natural skin texture and fine grain.

Ask for changes your photo can support. A sharp, front-facing head-and-shoulders photo supports a chest-up portrait in new light. A dark, tilted selfie is a harder start for a bright studio look. If your source already has window light, keep that direction instead of inventing a new one.

Do not ask to change what you want to keep. If you already like your sweater, say that it stays. The fewer changes you request, the easier it is to keep your face.

Run the prompt, then compare the result with your original. Change one part only, such as the light or the sweater, and run it again. If the result gets worse, go back to the previous version. Keep a short note of each version so you can return to the one that worked.

Tools read the same words differently, and one prompt can give a different image on each run. No prompt can promise a specific result. FrameGloss's own examples were made with Nano Banana 2, a Google Gemini image model, so wording that suits one tool may need adjusting in yours. Test in your own AI image tool and trust what you see.