DEV Community: Miriam Alonso

Best AI Headshot Generators Ranked by Likeness: 15 Tested on One Face

Miriam Alonso — Mon, 07 Sep 2026 10:16:34 +0000

We ran the same eight photographs of one man through the fifteen tools competing for "best AI headshot generator" and scored 57 selected outputs. The interesting result is not the ranking. It is that the failures all point the same way.

Disclosure. No affiliate links; every tool link goes to that tool's own homepage and nothing here earns a commission. The input photographs are of Ricardo Ghekiere, who founded BetterPic, one of the fifteen tools, and I work with BetterPic. It scored highest. The scoring and every image are published so the claim is checkable.

The subject

Fifty-something. High-volume grey hair. Full cheeks. A smile that is slightly uneven on one side.

Eleven of the fifteen tools returned a man who was younger, thinner-faced, or darker-haired. Often all three.

Why "random error" does not explain it

If these were noise, the errors would be symmetric. Some outputs would age him, widen the jaw, add grey. Across 57 images, essentially none did.

Every likeness failure we logged moved in one of these directions:

Grey removed or reduced. Headshot.kiwi in all 3 selected frames, HeadshotsByAI in all 4, Dreamwave, ProPhotos and Aragon in subsets.
Face narrowed. HeadshotPro in all 6 of its frames, at a consistent magnitude.
Hair volume inflated. The inverse, and rarer: PortraitPal returned 2 oversized hairstyles, Aragon 2 more.
Subject replaced. TryItOn AI returned one selected output that was a different person.

A one-directional error across eleven independent products, each with its own pipeline, is not eleven coincident bugs. It is a shared prior.

The mechanism, stated as a hypothesis

These products are personalisation layers — LoRA or DreamBooth-style fine-tunes of a few hundred to a few thousand steps — sitting on a general text-to-image base. You supply 8 to 15 photos; the fine-tune learns an identity token; generation is then conditioned on that token plus a prompt like professional corporate headshot, studio lighting.

That prompt is doing more work than the identity token. Whatever "professional headshot" means inside the base model was learned from stock photography, LinkedIn-adjacent corporate imagery and marketing shots — a distribution that skews young, symmetric and conventionally groomed. Where the subject agrees with that prior, the fine-tune has nothing to fight. Where the subject disagrees — grey hair, a fuller face — the prior competes with the identity token, and on a light fine-tune the prior wins.

That predicts exactly what we measured: the error is not "the model got the face wrong", it is "the model pulled the face toward the mean of its training distribution".

I am labelling this a hypothesis rather than a finding because our test measures outputs, not weights. It is falsifiable, though, and cheaply: undertrain and overtrain the same LoRA on the same subject and watch whether likeness degrades toward the prior specifically, rather than degrading in arbitrary directions.

Headshot.kiwi, where the prior wins outright.

HeadshotPro, the same pull at lower amplitude and applied uniformly.

The corollary nobody markets

Realism and likeness come apart, and every comparison that reports one number hides it.

Tool	Realism	Likeness
Headshot.kiwi	7	2.5
HeadshotsByAI	6	2
The Multiverse AI	7	4.5
BetterPic	8.5	9

Headshot.kiwi produces genuinely convincing photographs — good directional light, real skin texture, plausible depth-of-field falloff — of somebody else. Scored as "quality", it looks mid-table. Scored on whether it returned the person you uploaded, it is near the bottom.

This is why our rubric weights likeness at 40% and realism at 30%. A model that has learned to render faces beautifully and to render this face is doing two separable things, and only one of them is what the customer bought.

The falsifiable prediction

If the mechanism above is right, the failure rate should scale with distance from the base model's prior. Grey hair is one axis. The others, which this test did not cover, are the ones already reported anecdotally across the category: tightly curled hair, darker skin tones, and strong prescription glasses.

That is the obvious next test, and it is more useful than another ranking. One subject tells you about one subject — this whole exercise is a single face, one skin tone, one hair type, no glasses, and I would not generalise it further than the mechanism.

What came out on top, and where it slipped

BetterPic scored 9 on likeness, the highest in the test, and 8.6 overall. The frame we scored highest keeps the grey, the volume, the skin texture and the uneven smile at once.

It is worth naming where it slipped, because it slipped in the same direction as everything else: its grey-suit frame — the formal one, the one you would actually use — is the weakest of its four. The hair is combed flat and the grey is cut back to a streak at the temple. The prior does not disappear at the top of the table. It just gets outvoted more often.

Every score, all 57 outputs and the 8 inputs: the full gallery.

Best AI Headshot Generator 2026: We Tested 15 on the Same Face

Miriam Alonso — Mon, 07 Sep 2026 10:16:14 +0000

Most comparisons of the best AI headshot generator are a list of prices and a paragraph of vibes. This one is a measurement, so the first thing to publish is the instrument.

Disclosure, because it changes how you should read this. There are no affiliate links here and nothing on this page earns a commission — every link goes to the tool's own homepage. The eight input photos are of Ricardo Ghekiere, who founded BetterPic, which is one of the fifteen tools tested, and I work with BetterPic. BetterPic came first. The rubric is published below precisely so you can check that result rather than take it.

The setup

One subject. Eight input photos, identical for all fifteen tools. We paid full price on every one — no press access, no free tier where a paid tier existed.

From each tool we selected 3 to 6 output frames, picking from that tool's own style library. Fifty-seven selected results in total.

The rubric

Four criteria, weighted:

Criterion	Weight	What it measures
Likeness	40%	Is this recognisably the same person?
Realism	30%	Does it read as a photograph rather than a render?
Usable shots	20%	How many frames could you actually put on a profile?
Price per usable image	10%	Cost divided by the frames that survived

Likeness carries the most weight because it is the failure nobody markets and everybody notices. A generator that produces a beautiful portrait of somebody else has failed at the only job it had.

Every score

#	Tool	Score	Likeness	Realism	Usable	$/usable
1	BetterPic	8.6	9	8.5	8	$8.75
2	Secta AI	8.1	8.5	8.5	7.5	$16.33
3	HeadshotPro	8.0	7	8.5	8.5	$4.83
4	Headshots.com	6.6	6.5	8	6	$30.00
5	Dreamwave AI	6.3	7	5.5	6.5	—
6	PortraitPal	5.5	5.5	6	5	—
7	Aragon AI	5.3	5.5	5	5	$17.50
8	ProPhotos AI	5.2	5.5	6.5	3.5	$39.00
9	The Multiverse AI	5.0	4.5	7	3.5	—
10	StudioShot	4.7	4.5	6	3.5	—
11	AI SuitUp	4.0	3.5	5.5	2.5	$27.00
12	TryItOn AI	3.9	2.5	5.5	3	$17.00
13	InstaHeadshots	3.6	4	4	3	—
14	Headshot.kiwi	3.4	2.5	7	1	—
15	HeadshotsByAI	2.9	2	6	1	—

The interesting part: realism and likeness are not correlated

Look at rows 14 and 15. Headshot.kiwi scores 7 on realism and 2.5 on likeness. HeadshotsByAI scores 6 and 2. Both produce photographs that would pass as real if you had never met the subject — good low-key lighting, natural skin, believable depth of field — of a man who is not him.

Headshot.kiwi, scored 7 on realism and 2.5 on likeness. Good light, real skin, wrong man.

That is the single most useful finding in the test, and it is invisible in every comparison that scores "quality" as one number. These models are not failing to generate a convincing photograph. They are succeeding, and substituting the face while they do it.

The failure taxonomy

Across the eleven tools that scored below 7, the errors were not random. Four kept recurring:

1. Grey substitution. The subject has high-volume grey hair. Headshot.kiwi replaced it in all three selected frames with the same younger, darker, narrower version. HeadshotsByAI did the same across four. Dreamwave, ProPhotos and Aragon each did it in a subset.

2. Face narrowing. Full cheeks get slimmed. HeadshotPro did this in all six of its frames — consistently enough that it reads as a house style rather than a defect, which is why it still scored 7 on likeness and finished third.

HeadshotPro's narrowing, above — consistent enough across all six frames to read as a house style.

3. Hair overgrowth. The inverse error. PortraitPal produced two oversized hairstyles, one attached to a stretched face; Aragon produced two hair caricatures alongside two usable frames.

4. Wholesale replacement. TryItOn AI returned one selected image that had replaced the man entirely. Not a distortion — a different person.

What won, and where it lost

BetterPic took first place at 8.6 on the strength of likeness: 9 out of 10, the highest in the test. The navy t-shirt frame keeps the high grey hair, the skin texture and the uneven smile — the three things everything else flattened.

Where it lost: our closest likeness is the one wearing a t-shirt, which is not what most people want from a professional headshot. The grey-suit frame is the weakest of its four — hair combed flat, the grey cut back to a temple streak, jaw slightly narrower under formal light. And at $35 for four usable frames, $8.75 per image is nearly double HeadshotPro's $4.83.

If price per usable image is your metric rather than likeness, the ranking inverts and HeadshotPro wins outright. That is a legitimate way to weight it. It is not how we weighted it, and the table above lets you redo the arithmetic.

Caveats worth stating

Selection is the soft spot, twice. We chose each tool's style from its own library, then chose which 3 to 6 outputs to show. Both choices move the result. A different picker gets a different table.

Sample sizes differ. Secta AI returned three usable frames from three selected — perfect, and a thin basis for a judgement. HeadshotPro's six-for-six is a stronger claim on consistency than anything in the top two.

One subject. One face, one skin tone, one hair type. These tools fail unevenly, and most reliably on strong prescription glasses, tightly curled hair and darker skin tones. A fifteen-tool test on one man tells you about one man.

The rest of the images

All fifty-seven selected results are on the full version with every gallery, alongside the eight inputs.

The resolution maths behind AI headshots, and why most tiers are sized wrong

Miriam Alonso — Tue, 01 Sep 2026 09:36:38 +0000

Written by the BetterPic team. Disclosure: BetterPic is one of the tools named below. Vendor figures are their published numbers as of 1 September 2026 and are checkable on their own pricing pages.

Every AI headshot generator sells its tiers on image count and turnaround time. Almost none of them sell on output resolution, which is the one spec that determines whether the file is usable for the thing you bought it for. Here is the arithmetic, because once you have it the tier comparison stops being a marketing exercise.

The only formula you need

Pixels needed = target print size in inches x DPI.

Print work is 300 DPI. So an 8-inch print needs 2,400 pixels on the long edge. A 3.5-inch business card headshot needs 1,050. A conference badge at 2 inches needs 600.

Screen work is different and much less demanding, because you are sizing to CSS pixels and a device pixel ratio, not to physical inches. A LinkedIn profile photo renders at 400 x 400, and 800 x 800 when expanded. A Slack avatar is smaller still. On a 2x display you want double, so 1,600 pixels on the long edge covers essentially every screen use you will ever have.

That gives two clean thresholds: 1,600 px for anything on a screen, 2,400 px for anything that gets printed at a sensible size.

What the tools actually output

Aragon Basic: 896 x 1088. Executive: 1792 x 2176.
HeadshotPro: standard JPEG, with a 4K upscale available on the $59 tier.
BetterPic: 4K on every plan including the free one, plus TIFF export at 300 DPI.

Run those against the thresholds. Aragon's Basic at 1,088 on the long edge does not clear the screen threshold on a 2x display, let alone print — it is about three inches at 300 DPI. Its Executive tier at 2,176 clears screens comfortably and lands just short of an 8-inch print. HeadshotPro's standard output clears normal screen use, and the print case is what the paid upscale is for.

The point is not that one vendor is better. It is that the cheapest tier of at least one popular tool is sized below what a modern phone screen will show, and no listicle mentions it because listicles compare price and image count.

A warning about the word "4K"

"4K" in this market means roughly 4,000 pixels on the long edge, and it is doing a lot of work in marketing copy without being a standard. Two things to keep in mind.

First, an upscale is not the same as a native render. Upscaling a 1,024-pixel generation to 4,096 gives you a 4,096-pixel file, not four times the detail. It can look better — modern upscalers are genuinely good at faces — but it will not recover detail in the places these models are weakest, which is hair edges and the boundary where glasses frames meet the temple. Look at those two regions at 100% before deciding an upscale worked.

Second, JPEG at high resolution is still JPEG. If the output is going into a print workflow, you want a format without generation loss, which is what TIFF export is for and why it appears on spec sheets at all.

Where this leaves the choice

If everything you produce lives on a screen, resolution is a solved problem on every tool and you should choose on turnaround time and refund policy instead. Aragon and HeadshotPro turn around in about 30 minutes; BetterPic takes one to two hours, so it is the slowest of the three. All three publish one, and the conditions differ: HeadshotPro's has no expiry and no questions asked, Aragon's requires that you have not downloaded, and BetterPic's is 7 days and conditional on the model not yet having been trained on your photos. HeadshotPro's is the strongest of the three.

If any of it gets printed, resolution is the first filter and it eliminates the cheap tiers quickly. BetterPic is the one that outputs 4K natively at every price point with TIFF at 300 DPI, which is the reason to look at it here; Aragon's Executive tier is the other option that clears the bar.

And regardless of tool: check the output on your own face rather than the sample gallery. These models fail unevenly, most reliably on strong prescription glasses, tightly curled hair, and darker skin tones. That variance is larger than the variance between vendors, and it is not visible in any spec sheet.

Further reading

There is an index of everything we have published.

More on this subject, grouped by topic: Everything published here, by subject

Disclosure: the tool I work on is one of the few that ships 4K AI headshots.

Rolling out headshots for a whole company is a logistics problem, not a photography one

Miriam Alonso — Tue, 01 Sep 2026 08:11:22 +0000

Ask what makes company headshots difficult and most answers describe the shoot: lighting, a backdrop, someone who knows what they are doing behind the camera. That part is solved and has been for a century.

What actually breaks a rollout is everything around it.

The four problems nobody budgets for

Scheduling. A studio day for sixty people means sixty calendar slots, and the ones who miss theirs are the ones you most need. Remote staff cannot attend at all. In practice a booked day covers perhaps seventy per cent of the company, and the remainder drags on for weeks.

The March problem. You finish the rollout in January and hire eleven people by spring. Their photos come from a different session, in different light, against a different wall. Six months later the team page is visibly two cohorts. This is the failure mode that makes companies quietly abandon the whole idea.

Consent and likeness. People have opinions about their own face, and they are entitled to them. Someone will hate their photo and want it replaced. Someone will leave and want it removed. Someone will not want to be photographed at all. A process with no answer to these gets stuck at the first objection.

Uniformity across faces. Covered at length elsewhere, but briefly: optimising each portrait individually produces a worse page than applying one consistent treatment to all of them.

What changes when the constraint is not a shoot day

If portraits come from photos people already have, three of those four problems change shape. Scheduling stops being a shared date and becomes an individual task. The March problem becomes a matter of running the same treatment on one new person. Replacing an individual photo no longer requires reconvening anything.

Consent does not get easier and should not. It just stops being coupled to a logistics deadline, which is what usually turns it into a rushed conversation.

This is the whole argument for doing team headshots this way, and it is a workflow argument rather than a quality one. The organisation-wide version, including how the styles stay pinned across cohorts, is at BetterPic for teams.

What to check before committing

Ask for the same style applied to two people who look nothing alike, and look at them side by side. Uniformity across genuinely different faces is the thing that separates a coherent team page from forty images that happen to share a background.

Then ask what happens in March, and see whether the answer is a process or a shrug.

The individual case, where none of this logistics burden exists, is a different problem with different priorities: that is covered in the professional headshot guide.

Further reading

Related, on our comparison blog: Corporate headshots: what changes when it is fifty people, not one

There is an index of everything we have published.

More on this subject, grouped by topic: Everything published here, by subject

Qui écrit les comparatifs de cette catégorie

Miriam Alonso — Fri, 28 Aug 2026 10:18:35 +0000

Vous cherchez le meilleur générateur de portraits et vous tombez sur des classements. Presque tous sont du contenu d'affiliation : l'auteur touche une commission si vous achetez.

C'est la norme du secteur et cela n'a rien de déshonorant — notre programme d'affiliation est public, et ses conditions aussi. Ce qui compte, c'est l'effet sur la liste.

Ce qui est faussé, c'est la présélection

Pas le jugement sur un produit donné, mais qui y figure. Un fournisseur sans programme d'affiliation n'apparaît dans aucun comparatif, si bon soit-il. Vous ne lisez pas un classement des meilleurs outils, mais des meilleurs outils qui versent une commission.

Le second effet est plus discret : l'ordre suit souvent le montant de la commission. D'où dix articles différents affichant les trois mêmes noms dans le même ordre.

Comment reconnaître un vrai test

Précise-t-il les photos de départ ? Quinze du même après-midi et quinze réparties sur trois ans donnent des résultats sans rapport.

Montre-t-il un échec ? Tous ces systèmes échouent sur certains visages : lunettes à forte correction, cheveux très bouclés, peaux foncées, asymétries marquées. Un comparatif sans un seul essai raté n'a pas testé, il a trié.

Porte-t-il une date ? Les modèles changent tous les quelques mois et les prix ont fortement baissé — les chiffres du marché montrent le mouvement. Sans date, un test décrit un autre marché.

Où lire

Les générateurs dans leur ensemble pour la catégorie, et les générateurs de portrait pour ce qui sort du contexte professionnel.

Pour la photo de profil les critères diffèrent : les outils pour le réseau professionnel visent le sérieux, les outils de photo de profil en général visent la reconnaissance sur plusieurs plateformes. Et en contexte d'entreprise, ce qui compte n'est plus la qualité d'une image mais la cohérence entre quarante.

Les incertitudes récurrentes sont dans les questions fréquentes.

Le test dont personne ne tire commission

Passez votre propre visage dans deux versions gratuites. Vingt minutes de preuve personnelle valent mieux que dix articles d'opinion.

Quién escribe las comparativas de esta categoría

Miriam Alonso — Fri, 28 Aug 2026 10:03:16 +0000

Buscas el mejor generador de retratos y aterrizas en listas. Casi todas son contenido de afiliación: quien las escribe cobra si compras.

Es lo normal del sector y no tiene nada de deshonroso — nuestro programa de afiliados está publicado, y sus condiciones también. Lo que importa es el efecto que tiene sobre la lista.

Lo distorsionado es la preselección

No el juicio sobre un producto concreto, sino quién aparece siquiera. Un proveedor sin programa de afiliados no sale en ninguna comparativa, por bueno que sea. No lees un ranking de las mejores herramientas, lees uno de las mejores herramientas que pagan comisión.

El segundo efecto es más discreto: el orden suele seguir el importe de la comisión. Por eso diez artículos distintos enseñan los mismos tres nombres en el mismo orden.

Cómo se reconoce una prueba de verdad

¿Dice de dónde salieron las fotos? Quince de una misma tarde y quince repartidas en tres años dan resultados incomparables.

¿Enseña algún fallo? Todos estos sistemas fallan con ciertas caras: gafas de graduación fuerte, pelo muy rizado, tonos de piel oscuros, asimetrías marcadas. Una comparativa sin una sola tirada mala no ha probado, ha seleccionado.

¿Lleva fecha? Los modelos cambian cada pocos meses y los precios han caído mucho — las cifras del mercado muestran el movimiento. Sin fecha, una prueba describe otro mercado.

Dónde leer

Los generadores en conjunto para la categoría, y las aplicaciones para todo lo que va en el móvil — ahí lo decisivo no es la calidad de imagen sino qué hacen con tus subidas después.

Para la foto de perfil los criterios cambian: las herramientas de foto de perfil buscan que te reconozcan en varias plataformas, mientras que lo que aplica en la red profesional busca legibilidad en una lista. No es lo mismo.

Las dudas que se repiten están en las preguntas frecuentes.

La prueba por la que nadie cobra

Pasa tu propia cara por dos capas gratuitas. Veinte minutos de evidencia propia ganan a diez artículos de opinión ajena.

Diagnose before you reshoot

Miriam Alonso — Fri, 28 Aug 2026 09:48:01 +0000

The reflex when a profile photograph disappoints is to take another one. That is usually the most expensive route to a five-minute problem.

What is actually wrong, most of the time

The crop. Everything below the collarbone is wasted area at the size these images are displayed, and a head occupying a quarter of the frame simply stops being identifiable.

Correcting it needs no camera, no session and no money. It is worth exhausting before considering anything else, and it resolves roughly half of all cases.

The second most common finding

The background. Dark hair against a dark wall gives a silhouette with no face in it once scaled down — invisible at full size, total in a list. The fix is not white but separation from your hair colour.

The third

Light direction. Strong side light is flattering at full size and turns half the face into a grey area when scaled down. Flatter and more frontal is duller in large and much more legible in small.

None of the three depends on the camera, which is why buying a better one does not help.

The brief for this destination

The core requirements come down to those three, and the professional version covers the conventions, which are narrower than for portraits generally.

Taking it yourself is entirely feasible. The error to avoid is distance: at arm's length perspective enlarges your nose, and two metres back with zoom removes the effect completely.

What the photograph can claim

Less than the marketing says and more than sceptics allow. It adds no competence. It prevents someone briefly willing to look from scrolling past — a modest claim and a defensible one.

Reading further

The collected material on this network covers the rest.

The one-minute test

Scale your image down to the size it is actually displayed at and set it beside the original. If you cannot tell where your head ends, that is the version almost everyone sees.

Almost nobody runs this test, and it is the only one that reproduces the real viewing conditions.

Further reading

There is an index of everything we have published.

The background decides more of your profile photo than your face does

Miriam Alonso — Fri, 28 Aug 2026 09:32:48 +0000

Ask what is wrong with a profile photograph and people talk about their expression. Look at the actual failures and most of them are behind the person.

Why the background wins

At full size a dark background behind dark hair looks considered. Scale it down and the head loses its edge entirely — a silhouette with features somewhere inside and no boundary.

Undetectable in the conditions used to choose, total in the conditions where the image is seen. That is why it survives so long unnoticed.

The fix is not white

It is separation from your own hair colour, a different colour for different people. Light hair needs a mid or dark background; dark hair needs a lighter one. Blanket advice fails for about half of people and gives the rest a cut-out, floating look.

Which backgrounds work in this context treats it as a contrast decision rather than a taste one, which is the correct frame.

The second-order variable

Distance. A metre or more between you and the wall throws the background slightly out of focus and removes texture competing with your face. Standing against the wall does the opposite, and is what most people do.

The brief for this specific destination

The requirements for a professional network start from the viewing conditions rather than from photographic ideals: one image, seen small, in a list of eleven other faces, with about a second of attention.

Three things survive and none involves a camera: hair-background contrast, a tight crop, frontal light.

Where the photograph sits in the profile

One field among several and not the highest-leverage one. Improving the profile with the image as one input treats it in that proportion — the headline text does more work in most searches.

If you are generating

The generator for this use covers the mechanics, and the case for reusing one image across a profile and a website covers what to do with the output. Recognition across platforms is worth more than optimising any single placement.

Before spending anything

Free tools handle crop, size and background replacement, covering about half of real cases at no cost.

The test

Scale it down. If your head has no edge, it is the background — and it always was.

Further reading

Related, on our comparison blog: Your LinkedIn profile picture is seen at 48 pixels

There is an index of everything we have published.

If the background is the problem rather than the photograph, a free profile picture editor.

More on this, on our Quora Space: why the background's job is contrast rather than taste

Where to get one, ranked by how much trouble it is

Miriam Alonso — Fri, 28 Aug 2026 09:17:38 +0000

"Where do I get a professional headshot" has four real answers. Which is right depends on something rarely asked first: how many do you need, and by when.

One: fix the photograph you already have

The only free option and the most underrated. Roughly half of the portraits people call bad are decent photographs with the wrong crop — too much shoulder, head too small, everything below the collarbone eating the frame.

Five minutes, no new photograph. Rule it out first, because if it works nothing else is needed.

Two: photograph yourself properly

Also free, and better than expected with three things right.

Distance is the one people miss. At arm's length, perspective enlarges whatever is nearest the lens and shrinks what is furthest — a bigger nose and smaller ears, which is geometry rather than the camera being unkind. Two metres back with zoom removes it entirely.

Then a window rather than the ceiling light, and a metre between you and the wall.

Three: generate a set

The case is repeatability rather than quality. Where results actually look real is the specific version of the question, and one walkthrough shows the process end to end.

The preparation that decides the outcome is not the tool: source photographs from different days, in different clothes, in different light. Fifteen from one afternoon in one shirt teaches the system that the shirt is you.

Four: book a photographer

Still right for one important portrait, and the reason is not equipment. It is a person watching you and saying your chin is slightly too high — the actual product of a session, and nothing reproduces it.

The full comparison of routes covers costs, and the figure that surprises people is not the fee but the two or three hours of travel, session and selection that appear on no quote.

How to choose in one question

One photograph or forty. For one, a photographer is defensible at almost any price. For forty across people who joined in different months, it is a scheduling problem, and a session does not solve scheduling.

The check that costs nothing

Scale your current photograph to profile size. If the outline of your head survives and your eyes are visible, you may not need any of the four.

Further reading

There is an index of everything we have published.

Disclosure: I work on one of the paid options in that ranking — BetterPic.

The specifications decide more than the advice does

Miriam Alonso — Fri, 28 Aug 2026 09:02:28 +0000

Most guidance about professional portraits is about how to look. What decides whether an image works is three measurements, and they are dull enough that almost nobody checks them.

The display size

Not the file size — the size it is actually seen at. In a list, roughly forty-eight pixels. On a conference badge, about fifteen millimetres. On a business card, less.

At those sizes skin texture, eyebrows and fine gradients are gone, leaving the outline of the head and two or three masses of light and dark.

The crop ratio

Everything below the collarbone is wasted area. The most common fixable problem in the whole subject, and it needs no reshoot.

The contrast between hair and background

Invisible until the image is small, then total. The fix is separation from your hair colour, which is a different colour for different people — which is why blanket advice to use white fails for about half of people.

Then the conventions

The general treatment covers those, and it is worth reading against the three numbers rather than instead of them. Advice about expression and wardrobe is real but second order; geometry decides first.

Going deeper by subject

The material splits into genuinely different problems and it is worth landing on the right one. The index lists them.

The tools section is about choosing software, where the criterion that matters is not sample quality but whether the output still looks like the specific person.

The industry section covers profession-specific conventions, which genuinely contradict each other — medicine wants warmth, law wants restraint, and no single photograph satisfies both.

The professional portraits section collects the rest.

The thing you can give someone else

A gift card rather than a finished image, because a portrait is a statement about how someone wants to be seen and that cannot be made on their behalf. Assigning someone a face lands as an obligation; giving them the ability does not.

The test

Scale your photograph to the size it is displayed at and set it beside the original. If you cannot tell where your head ends, that is the version almost everyone sees, and no resolution changes it.

Further reading

Related, on our comparison blog: Realtor headshots: the sign, the card and the profile need different files

There is an index of everything we have published.

What generated portraits are genuinely good at

Miriam Alonso — Fri, 28 Aug 2026 08:47:18 +0000

The public argument is framed as replacement. Wrong frame, and both answers to it are wrong. Here is where the technology actually earns its place.

The fortieth person

A session is an event. A workforce is a flow. Someone joins in March and does not match the January set, and eighteen months later the team page is mixed again.

Same crop, same background, same colour temperature on the fortieth face and the four hundredth — a different technical problem from making one good image, and the strongest case in the category. Companies with real brand specifications hit a stricter version of it: a specified background value, a specified crop ratio, an aspect that has to match existing assets.

Volume against a deadline

Sports leagues are the clearest example: hundreds of players across dozens of clubs, all needed before a season opens. The constraint is calendars, not cameras. The wider argument for that use turns entirely on consistency across people.

What it cannot do

Tell you what is wrong. A photographer looks at you and says your chin is slightly too high. That feedback loop is the real product of a session, and no generation step reproduces it. Where the two sit relative to each other makes the point more fairly than most vendor pages.

Where it fails, consistently

Faces further from the average: strong glasses, tightly curled hair, darker skin tones, marked asymmetry. And the failure is not obvious — the image looks real, it just looks like someone who is almost you.

The five things that actually go wrong

The common style problems are worth reading against that, because four of the five are free to fix: no edge between hair and background, wrong crop, side light, and the wrong photograph for the destination.

The fifth is the source photographs, and it applies only if you are generating.

How the process actually works

The plain explanation strips the marketing. The system learns what is constant about you and separates it from what is circumstantial — which is why fifteen images from one afternoon in one shirt teach it that the shirt is part of your face.

That single input decision moves the result more than switching vendors does.

Where the category stands

The market research update has numbers rather than opinions: quality plateaued, prices fell, and any comparison written before roughly 2025 describes prices that no longer exist.

Further reading

More on this: AI headshots for job applications: where they help and where they hurt

More on this subject, grouped by topic: Everything published here, by subject

Disclosure: I work on BetterPic's AI headshot generator.

What "professional grade" actually adds

Miriam Alonso — Thu, 27 Aug 2026 12:51:08 +0000

Paid tiers in this category advertise quality. Quality is not what separates them, because the free tier draws from the same distribution.

What actually separates them

Repeatability. One good image out of forty attempts is luck. Forty usable out of forty is a product. That distinction only appears if you generate more than you need and count, which is tedious and therefore rarely done in reviews.

Consistency across people. The moment it is a workforce, the requirement stops being quality and becomes uniformity — the same crop, background and colour temperature across faces with different hair and colouring. A different technical problem, and a harder one.

Usage rights that hold. Irrelevant for a personal profile. Not irrelevant for advertising.

The professional-grade version is built around those, and how the generators work covers the mechanism underneath without the marketing layer.

The case with actual specifications

Residency and medical application photographs come with stated requirements — often a background colour and a crop — which turns a taste decision into a compliance one.

Check the rules before producing anything. A rejected photograph costs an application cycle, and the requirements are usually published somewhere unhelpful.

What to do with the output

More than one placement. Using one portrait across an online presence is realistic about the limits, and the underrated benefit is recognition: the same face in five places is worth more than five better faces in five places, and it costs nothing.

Why the prices moved

The cost of running these models fell sharply, and the category went from expensive to nearly free at entry level. The market numbers show the movement.

The consequence for a buyer is concrete: any comparison written before roughly 2025 describes prices that no longer exist, so check the date before the ranking.

The thing that decides the result regardless of tier

Source photographs from different days, in different clothes, in different light. Fifteen images from one afternoon in one shirt teach the system that the shirt is part of your face.

No paid tier rescues a bad input set, and no free tier is held back by a good one.

Further reading

A longer piece on this: What "studio quality" actually describes

Disclosure: I work on one of the paid tools this compares, BetterPic.