Headshot guide
AI headshot prompts for ChatGPT and Gemini
These AI headshot prompts turn one everyday photo into a studio-style headshot in ChatGPT or Gemini (Nano Banana). We tested them with GPT Image 2 and Nano Banana Pro on 2026-10-04 and rewrote the lines that let the models drift.
UpdatedA professional headshot prompt to copy and paste
Open ChatGPT or Gemini, attach a clear photo of yourself and paste the prompt below. It makes a gray studio headshot in a navy blazer; the table further down has other looks.Use the photo I've attached only to learn what I look like. Make a new professional headshot of me, as if a portrait photographer took it today. Don't copy the background, clothes, lighting or camera angle from my photo. Keep me exactly as I am: the same face shape and width, jawline, eyes and eye color, eyebrows, nose, lips, teeth, skin tone, freckles, moles, wrinkles, hairline, hair color and texture, facial hair, and my glasses if I wear them. Keep my age, weight and build. Don't slim my face, smooth my skin, whiten my teeth or make me look younger. Pose: shoulders turned slightly away from the camera, face turned back to the lens, chin a little forward and down, a relaxed, natural smile. Framing: head and shoulders, cropped at mid-chest, eyes about a third of the way down the frame, a little space above my hair. Clothes: a navy blazer over a plain white shirt. Background: plain mid-gray studio backdrop, a little lighter behind my head. Light: a large soft light in front of me and slightly to one side, soft fill from the other side, a small catchlight in each eye. Camera: 85mm lens at f/4, from about two meters away at eye level, eyes sharp, background softly out of focus. Real skin with visible pores and fine lines, natural color. No airbrushing, no plastic look, no text. Vertical 4:5 format.If you have two or three photos of yourself, attach them all and change the first sentence to: All the photos I've attached are of me. Use them only to learn what I look like. In our test, extra photos helped Gemini more than any wording change.If the result makes you slimmer, younger or smoother than you are, add a plain sentence about yourself after the third sentence, such as: About me: I'm a woman in my mid-forties with a full, round face and a heavier build. Name anything a model might tidy away: age, weight, face shape, skin tone, hair texture, a scar, a gap in your teeth.What each part of the prompt does
| Part of the prompt | Why it's there |
|---|---|
| Use the photo only to learn what I look like, and don't copy its background, clothes, light or angle | Edit models tend to paste the selfie onto new clothes, keeping its head tilt, expression and close-up distortion. This line asks for a new photo instead. |
| Keep me exactly as I am, followed by the list of features | Image models beautify by default: smoother skin, whiter teeth, bigger eyes, narrower faces. Naming each feature and each kind of beautifying leaves less room to drift. |
| Shoulders turned slightly away, face back to the lens, chin a little forward and down | The standard headshot pose. Square shoulders look boxy; chin forward and down defines the jawline. |
| Head and shoulders, eyes about a third of the way down | Keeps your face big enough to read as a thumbnail. With only "head and shoulders", GPT Image 2 framed much looser. |
| A large soft light slightly to one side, soft fill, a catchlight in each eye | Loop lighting, the usual corporate setup. A small shadow beside the nose gives the face shape, and catchlights keep the eyes alive. |
| 85mm lens at f/4, about two meters away, eye level | Photographers stand back with a longer lens so faces keep their real proportions. At f/4 from two meters about 13 cm of depth is sharp: both eyes and the nose, not the backdrop. |
| Real skin with visible pores and fine lines, no airbrushing | Waxy, plastic skin is a common complaint about AI headshots. Asking for texture works better than only forbidding smoothness. |
| Vertical 4:5 format | Gemini returned exactly the shape we wrote, 4:5 or 1:1. 4:5 suits bios and websites; square suits LinkedIn and avatars. |
Prompts for other looks: LinkedIn, outdoor and creative
Keep the rest of the prompt and replace the Clothes, Background and Light sentences with one row. We tested all three.| Look | Clothes | Background | Light |
|---|---|---|---|
| Bright office | a dark blazer over a light shirt | a bright modern office far behind me, softly out of focus, no readable screens or signs | soft window light from one side, gentle fill from the other |
| Outdoor | a casual blazer over a plain crewneck | trees and greenery well behind me, softly out of focus | soft daylight in open shade, with a little warm sunlight from behind catching my hair |
| Creative | a dark knit sweater | a warm, textured plaster wall a few meters behind me | soft window light from one side, letting the far side of my face fall into gentle shadow |
LinkedIn headshot prompt
LinkedIn shows your photo as a circle cut from a square, so change two more sentences. Replace the Framing sentence with: Framing: square head-and-shoulders crop, my head centered with clear space above my hair and below my chin, so a circular crop keeps my whole head. Replace the last sentence with: Square 1:1 format. Use the bright office row or keep the gray studio.With this framing our test headshots had the head, from hair to chin, at 54% to 63% of the frame's height, and the whole head stayed inside the circle. That's about right for a photo people mostly see small.How to get an AI headshot that looks like you
- Start from a good photo. Face a window in daylight, with no sunglasses, no filter and nobody else in the frame. A photo someone else takes from a few steps back beats an arm's-length selfie.
- Give Gemini two or three photos: one straight on, one with your head turned about three-quarters, and one taken by someone else. In our test, three photos fixed most of the face narrowing we saw with one.
- Tell ChatGPT what it must not change. One About me sentence kept a round face and heavier build that GPT Image 2 had slimmed. Extra photos didn't fix that.
- Make a few tries and pick. Each run comes out a little different, so generate three and keep the one that someone who knows you recognizes fastest.
- Check it at full size and at thumbnail size. Look at eye color, the hairline, moles and scars, and your teeth. If a friend would hesitate, don't use it.
Fix-it prompt for when it doesn't look like you
Start a new chat, attach your original photo first and the headshot second, then paste this with the brackets filled in:The first image is my original photo. The second is the headshot you made, and it doesn't look enough like me: [say what's off]. Fix the face so it matches my photo exactly: my real face shape, features, skin tone and age. Keep the clothes, background, light and framing of the headshot as they are. Don't smooth, slim or beautify anything.Be specific about what's off: my face is wider and rounder, my nose is broader, my eyes are further apart, my hair is thinner on top, I'm older than that, my skin is darker and warmer.In our test it fixed age: a 77-year-old who had come back looking younger got his deep wrinkles, drooping eyelids and gray stubble back, though he lost his smile along the way, so add "keep my smile" if you want it. It barely changed a face that had come out too narrow. If two rounds of fixes don't get there, start again with two or three photos and an About me sentence.What happened when we tested these prompts
We ran these prompts on 2026-10-04 with OpenAI's GPT Image 2 and Google's Nano Banana Pro (Gemini 3 Pro Image), calling the models directly, so the ChatGPT and Gemini apps may behave a little differently. The subjects were five synthetic people made for testing, aged about 24 to 77, so no real person's face was used. From everyday phone photos taken in cars and kitchens we made 22 headshots.- Every result looked like a real studio photograph. Whether it looked like the right person varied by model and by face.
- GPT Image 2 kept faces closest from a single photo, down to the beard, age spots, stubble and glasses, but it slimmed the woman with the heavier build.
- Nano Banana Pro rendered the most natural skin, with visible pores and fine lines. From one selfie it narrowed two faces and made the oldest man look younger, with fuller hair.
- Both kept glasses, gray hair and freckles, and Gemini followed the 4:5 and 1:1 format lines exactly.
The limits of DIY headshots in ChatGPT and Gemini
- You get one headshot per prompt, and each try drifts a little. The person in your outdoor shot may not quite match the person in your studio shot.
- A matching set is hard. Light, color and skin tone change from run to run, which shows on a team page or a speaker page with several photos.
- Both label what they make. Gemini adds an invisible SynthID watermark and C2PA Content Credentials to every image, plus a visible sparkle watermark that most people can switch off in its settings. ChatGPT images carry C2PA metadata and a SynthID watermark, with no visible mark. LinkedIn shows a C2PA icon on feed images that carry these credentials.