Short answer: to get a professional headshot from a selfie, attach a sharp, front-facing selfie in even light to an image model that edits photos (Nano Banana, GPT Image or Flux Kontext), and ask it to change only the clothes, the background and the lighting. Add one explicit rule: keep the face, bone structure, skin tone and hair exactly as in the photo. That rule is what keeps the result looking like you.
The headshot prompts in the Prompt Gallery are built this way. Here is the method behind them.
Which selfie should you start from?
The model can only keep what it can see. A better selfie does more for the result than a better prompt.
- Face the camera, eyes level with the lens. A low angle changes the shape of your jaw in every result.
- Even, soft light, such as a window in front of you. Hard side light makes the model guess half your face.
- Sharp and full size. No beauty filter, no heavy compression from a chat app.
- Nothing covering the face: no sunglasses, no hand on the chin, hair away from the eyes.
- Arm's length or further. A phone held close to the face makes the nose look bigger, and the model keeps that.
If you have two or three good photos from different angles, attach them all. The model learns your face from several views better than from one.
Which tool should you use?
Text-only models cannot know your face, so you need one that takes a reference image:
| Tool | How to attach the selfie | Notes |
|---|---|---|
| Nano Banana (Gemini) | Upload it in the chat | Strong at keeping identity; ask for changes step by step |
| GPT Image (ChatGPT) | Upload it in the chat | Follows detailed briefs; say clearly what must not change |
| Flux Kontext | Image input plus instruction | Good for one precise edit at a time |
| Midjourney | Omni reference (--oref) | Keeps a likeness, but less exact for a real photo of you |
How do you stop the model changing your face?
Models like to "improve" faces: smoother skin, a slimmer jaw, whiter teeth, bigger eyes. For a headshot that people will compare with you in a meeting, that is a failure. Say so in the prompt, in its own sentence:
Preserve the exact facial identity, bone structure, skin tone and hair
of the reference with no stylisation of the face. Do not slim the face,
change the jawline, remove moles or alter the skin tone. Tidy, do not
replace."Tidy, do not replace" is the useful line. It allows the small things a retoucher would do (stray hairs, a shiny forehead) and rules out a new face.
Back it up in the negative prompt, or as a rule sentence if your tool has no negative field:
airbrushed plastic skin, altered jawline, slimmed face, whitened teeth,
removed moles, changed skin tone, uncanny smoothingWhat should you change instead?
Everything that makes a headshot look professional is outside the face. Describe it like a photographer's brief:
- Wardrobe: "a well-fitted plain knit in a muted colour" or "a tailored navy blazer over a white shirt". Plain colours age better than patterns.
- Background: "neutral grey studio backdrop" for a classic look, "softly blurred modern office" for a friendlier one, or two brand colours so the whole team matches.
- Light: "three-point studio lighting with a soft key" is the corporate standard. "Soft window light from the left" is warmer.
- Crop: "chest-up, room above the head, eyes on the upper third".
- Expression: "relaxed, confident, slight natural smile, direct eye contact".
Put together, the core of the gallery's Studio Headshot prompt reads:
Clean corporate studio headshot of the subject from the reference photo
in a tailored blazer, a neutral grey background, three-point softbox
lighting, tack-sharp focus, professional retouching, 85mm.How do you get several options at once?
One result is a coin toss. Ask for a grid instead: "four variations in a 2×2 grid, each the same chest-up crop, with a slightly different head angle and expression; same wardrobe, backdrop and lighting throughout." You then pick the best panel and ask for it again at full size. The Brand Headshot Generator works this way, with a backdrop in two brand colours so a whole team's photos match.
What size does a headshot need?
- LinkedIn and most profile pictures are shown as a circle. Use a square (1:1) crop with the face centred and margin around the head. The Profile Picture Pack makes six square variations that still read at 32 pixels.
- Company website, speaker page, press kit: 4:5 portrait is the usual shape.
- Slides and email signatures: square, small, high contrast against the background.
What should you check before you use it?
Look at the result next to your selfie, at full size:
- Would a colleague recognise you straight away?
- Are your eye colour, hairline, moles and glasses right?
- Do the hands, collar and ears look natural?
- Does the light on the face match the light on the background?
If something drifted, do not start over. Ask for one fix: "Keep everything, but restore the jawline from the reference photo."
A headshot represents you to people who will meet you. Use one that looks like you on a good day, not like someone else, and follow your employer's or platform's rules on AI images.
Where can you check this?
What is the short version?
- Start from a sharp, front-facing selfie in soft, even light.
- Use a model that edits photos: Nano Banana, GPT Image or Flux Kontext.
- Add the identity rule: keep the face exactly, tidy but do not replace.
- Change only the wardrobe, background, light and crop, and ask for a 2×2 grid of options.
The Prompt Gallery has tested headshot, profile picture and portrait prompts with the identity rule already written in. The course lesson on turning your own photo into a portrait or art style walks through it step by step, and for a character that is not you, see how to keep an AI character consistent.
