Free AI Headshot Prompt Generator (Professional Presets)

TL;DR: An AI headshot prompt generator turns a handful of choices — expression, wardrobe, background and light — into one platform-ready prompt, instead of a fixed list you edit by hand. This page gives you the five fields, the current settings for Midjourney V8.2, GPT Image 2, Ideogram 4.0 and Nano Banana Pro, and ten ready presets.

Most ai headshot prompt generator searches land on either a web-app landing page with no visible prompt, or a copy-paste list where someone else already decided your wardrobe and background. Our own list of 35 tested headshot prompts is a good version of the second kind, and it is still worth reading if you want finished prompts across professional, casual, creative and industry looks. This page is a different tool: five fields you fill in yourself, assembled into a prompt for whichever platform you are actually using. It also covers ground that list cannot, because it was written before Midjourney's Edit Model shipped and before Nano Banana Pro reached general availability — both matter for anyone trying to keep one face consistent across a batch of headshots.

What does an AI headshot prompt generator actually produce?

Text, not pixels. The generator's job is to turn 35-year-old marketing manager, confident, LinkedIn into a prompt that names a subject, an expression, a wardrobe, a background, a light source and a framing, in that order, then closes with whatever settings block the destination model needs. The image model does the rendering.

Five fields cover almost every headshot. Stack a sixth and a seventh and you start feeding the model contradictions it has to arbitrate on your behalf, like a warm smile and a serious expression in the same sentence — which is also the single most common reason a batch of otherwise-decent prompts comes back with dead, uncanny expressions.

Everything after that row is a platform choice, not a headshot choice: which model gets the five fields above, and what its own settings block looks like once they land there.

Which platform should the generator send your preset to?

The honest answer is that it depends on what you need the headshot to survive. Text-in-frame, resolution, and how hard it is to hold one face steady across ten variations are not the same axis, and the four platforms below split unevenly across them.

Four of those rows are the ones people get wrong from an out-of-date source, including our own live post 38.

Midjourney's face-lock parameter changed. --cref and --oref are stranded on V8.1 and V8.2 — Midjourney's own Parameter List now marks Omni Reference "(replaced by the Edit Model in V8.X)". Midjourney's Version article states plainly: "The current default Midjourney version is V8.2." The same article describes the Edit Model as the feature that "replaces Omni Reference, Character Reference, and the Retexture tool." Any headshot prompt still built around --cref is written for a version Midjourney no longer defaults to.

GPT Image 2 has no identity-lock field at all. It has no seed, no aspect_ratio, and no negative_prompt — the size parameter takes pixels or auto, and OpenAI's own limitations section says the model "may occasionally struggle to maintain visual consistency for recurring characters or brand elements across multiple generations." That is the vendor naming the exact failure mode a headshot batch runs into.

Ideogram 4.0's seed field is genuinely broken as a control. Its /v1/ideogram-v4/generate endpoint accepts no seed parameter, yet its response object still requires one back. You get a seed number in the reply that you were never allowed to set.

Ideogram's crop is not a guarantee, and Ideogram says so itself. Its own generation-settings documentation states plainly that "Ideogram may normalize your requested dimensions to a supported ratio, model size, or resolution tier." For a headshot destined for a specific crop, a LinkedIn circle, a badge template, treat the 1:1 or 4:5 request as a strong preference rather than a locked pixel count, and check the returned dimensions before you drop the file into a layout that assumes an exact size.

How do you keep the same face across an entire preset shoot?

Text alone gives you a plausible stranger wearing the wardrobe you described. To get your face across every preset in a set, anchor each generation to a real photo, and the mechanism for doing that is the one thing that differs most between these four platforms.

Midjourney: the Edit Model, not --cref. Attach an image in the Imagine bar (up to four), or add --edit to a Discord prompt followed by an image URL, and describe the change in plain instructions rather than the old keyword-stacking style. The Edit Model automatically matches the aspect ratio of your first upload unless you override it with --ar, and it works on V8.1 and V8.2 only.

Nano Banana Pro: reference images, budgeted by role. Google's Gemini API documentation gives gemini-3-pro-image up to 14 total reference images across three roles: up to 6 rendered with high fidelity as objects, up to 5 used to hold one character's identity steady, and up to 3 as style references. The mid-tier model, gemini-3.1-flash-image, gets a smaller character budget of 4 images and no style-reference slot at all, so check which model a tool is actually calling before assuming Pro's numbers apply. Our full guide to prompting Nano Banana Pro covers the resolution and pricing side of that same model in more depth than a headshot page needs.

Ideogram: one photo, auto-masked. Ideogram's Character feature takes a single reference image (by camera, upload, or your existing library), automatically masks the face and hair, and its own tip is exact: "A portrait-style image with a clear, well-lit face (preferably at a slight angle) yields the best results." Color Palette, Negative Prompt and Seed Number are unavailable while a Character Reference is active, and any Style Reference already applied gets disabled.

GPT Image 2: no dedicated slot, describe the person again. There is no equivalent lock. The closest lever is input_fidelity, and OpenAI's own docs say to omit it entirely for gpt-image-2 — "the API doesn’t allow changing it because the model processes every image input at high fidelity automatically." That removes one knob to worry about, but it does not replace an identity anchor.

One preset, four platforms

Same five fields, four settings blocks. This is the part a fixed prompt list cannot show you, because the point of a generator is that one combination of choices produces a different closing block depending on where it is going.

Fields: Late-30s marketing manager, confident closed-mouth smile, charcoal blazer over a cream top, softly blurred modern office, large window camera-left, 1:1 shoulders-up.

MIDJOURNEY V8.2
Late-30s marketing manager, confident closed-mouth smile, charcoal
blazer over a cream top, softly blurred modern office background,
large window light from camera-left, sharp focus on the eyes
--ar 1:1 --raw --stylize 100 --v 8.2

GPT IMAGE 2 (Responses API, image tool)
A photorealistic professional headshot of a woman in her late 30s,
confident closed-mouth smile, wearing a charcoal blazer over a cream
top, standing in a softly blurred modern office, lit by large window
light from her left side, sharp focus on the eyes.
size: "1024x1024"   quality: "high"

IDEOGRAM 4.0 (Style: Realistic)
Professional headshot, late-30s marketing manager with a confident
closed-mouth smile, charcoal blazer over a cream top, softly blurred
modern office behind her, large window light from the left, shallow
depth of field
Aspect ratio: 1:1

NANO BANANA PRO (gemini-3-pro-image)
{
  "prompt": "A photorealistic professional headshot of a woman in her late 30s with a confident closed-mouth smile, wearing a charcoal blazer over a cream top, in a softly blurred modern office lit by large window light from camera-left, sharp focus on the eyes.",
  "response_format": { "aspect_ratio": "1:1", "image_size": "2K" }
}

Notice what stayed fixed and what moved. The five fields are identical in every block. What changed is grammar: Midjourney reads flags after a comma-separated description, GPT Image 2 and Ideogram read a plain sentence, and Nano Banana Pro's own fields live outside the prompt string entirely, in response_format. A generator's actual job is holding the first part still while it rewrites the second part per destination.

10 copy-paste headshot presets, by field

Same structure as above, ten more combinations, spread across roles our own list of 35 does not cover and sent to whichever platform suits the job. Swap the bracketed values for your own details.

1. Software engineer, IC track (Midjourney V8.2)

[Late-20s software engineer], plain crewneck sweater, relaxed
half-smile, standing desk with a blurred monitor array behind, cool
overhead office light, no visible branding
--ar 1:1 --raw --stylize 80 --v 8.2

2. High school teacher (GPT Image 2)

A warm, approachable headshot of a [40s high school teacher], gentle
smile, cardigan over a collared shirt, softly blurred classroom
bookshelf behind, even daylight from a side window.
size: "1024x1024"   quality: "high"

3. Financial advisor (Midjourney V8.2)

[50s financial advisor], composed neutral expression, navy suit and
subtle patterned tie, dark wood office blurred behind, single warm
key light from upper left, trust-signal portrait
--ar 1:1 --raw --stylize 100 --v 8.2

4. Building architect (Ideogram 4.0, Style: Realistic)

Professional headshot of a [30s architect], confident slight smile,
black turtleneck, blurred drafting studio with a rolled blueprint on
a table behind her, soft diffused studio light
Aspect ratio: 4:5

5. Nonprofit program director (GPT Image 2)

A genuine, warm headshot of a [40s nonprofit program director],
open smile, patterned blouse under a soft blazer, blurred community
center interior behind, natural window light.
size: "1024x1536"   quality: "high"

6. Pharmacist (Midjourney V8.2)

[30s pharmacist in a white coat], warm reassuring expression,
name badge left blank, softly blurred pharmacy shelving behind,
even diffused overhead light, healthcare trust portrait
--ar 1:1 --raw --stylize 90 --v 8.2

7. Remote team directory shot (Nano Banana Pro)

{
  "prompt": "A clean, evenly lit headshot of a [30s remote employee], relaxed confident smile, simple crewneck shirt, plain neutral studio-grey background matching a team directory style, soft front-key light with no harsh shadow.",
  "response_format": { "aspect_ratio": "1:1", "image_size": "2K" }
}

8. Personal brand coach (Ideogram 4.0, Style: Realistic)

Professional-casual headshot of a [30s brand coach], direct engaged
eye contact, cream oversized blazer, blurred bright studio backdrop,
soft beauty-dish lighting, editorial but approachable
Aspect ratio: 4:5

9. University lecturer (Midjourney V8.2)

[40s university lecturer], thoughtful engaged expression, blazer over
a knit top, blurred lecture hall seating behind, warm side light,
academic portrait
--ar 3:4 --raw --stylize 120 --v 8.2

10. Insurance agent (GPT Image 2)

A friendly, trustworthy headshot of a [40s insurance agent], warm
smile showing teeth, smart-casual blazer, blurred small-office
interior behind, soft diffused daylight.
size: "1024x1024"   quality: "high"

Presets 3, 6 and 9 sit close to jobs our 35-prompt list already covers (banker, doctor, professor), so the wording and the platform choice above are deliberately different from that page rather than a restated copy of it. If you want the technique underneath these fields, the reasoning for why one light source beats three and how masking actually behaves per model, that is its own deep dive published alongside this page rather than a section here.

What settings actually make a headshot look photographic, not illustrated?

Three levers, and they are not the same lever on every platform.

Midjourney: --raw, still current on V8.2. Raw Mode is not a legacy parameter — it is the only current spelling. The old --style raw string no longer appears anywhere in Midjourney's own docs, and Discord's settings panel itself emits --raw. Pair it with a moderate --stylize (the Version article's own ladder runs Low 50, Medium 100, High 250) rather than pushing stylize high, which reintroduces the painterly look Raw Mode is turning off. --quality and --draft are both stranded on V8.1 and V8.2 — do not add either.

GPT Image 2 and Ideogram: quality: "high" and Style Realistic. Both are single-word switches rather than a numeric dial, which is simpler but gives you less room to nudge a result that is almost right.

Nano Banana Pro: nothing to turn off. Google's own documentation states that thinking is on for every generation on this model and cannot be disabled in the API, and the interim compositions it renders while thinking are not separately charged. If a render looks over-composed, changing the prompt is your only lever; there is no reasoning toggle to flip.

One habit works against you on every platform: stacking quality adjectives like 8k, hyper-realistic, award-winning instead of a real lighting and lens fact. FLUX.2's own guidance calls this exact pattern out as a leftover Stable Diffusion habit that current models read as noise rather than instruction — see our FLUX.2 generator if a headshot batch is really a uniform-lighting product-photography job in disguise, fifty employee photos shot the same way. One named light source and one named lens beat four superlatives, on every model in this table.

What still gives an AI headshot away?

Five tells, in roughly the order they show up:

1. Ears. Asymmetric in real faces, suspiciously matched in generated ones. If a preset is meant to anchor a real identity, a reference image fixes this faster than any prompt wording.
2. Jewelry and small hardware. A watch face, a lapel pin, a wedding band — these are exactly the small, structured details every vendor's own limitations language admits to struggling with.
3. Collar and lapel geometry. Suit lapels and shirt collars that don't quite meet where fabric actually folds. Naming a specific garment cut helps more than naming a fabric.
4. Background text. A blurred office with a legible sign or monitor is a giveaway at full resolution, even when the face is convincing.
5. A smile that reads as posed rather than held. Real portrait smiles have a slight asymmetry. Adding closed-mouth or naming one asymmetric detail, a slightly raised eyebrow, a head tilt, does more than another adjective describing warmth.

Do you need a paid plan to use this generator?

Prompt Architects does not render the headshot. It generates and stores the prompt, the way our product photography generator does for product shots: you fill in the fields, we assemble the wording and the settings block, and you send the result to whichever platform's playground, app, or API you already use.

Image prompt generation sits on the Pro plan and above, checked against our own pricing page in September 2026. The Free plan publishes five prompt enhancements per day, forever, on our FAQ page, checked the same day, which is enough to run one field combination through the enhancer and see what an assembled prompt looks like before anything costs money. Global Variables let you save your own subject and wardrobe once and reuse them across every preset above, so a ten-person team directory shoot is one saved template rather than ten hand-edited prompts.