AI headshot services charge $30–60 for a batch of images they generate from photos you upload. You can run the same class of model yourself for a few cents each, and your photos never leave your machine. The trade-off is that you do the culling and the prompt work — which is most of what you were paying for.
AI headshot services are a genuinely good product. You upload fifteen selfies, wait an hour, and get back forty images in various outfits and settings. For a lot of people that is the right purchase and this article will not talk you out of it.
But it is worth understanding what you are buying, because the model doing the work is one you can also run directly. The service is packaging: a fine-tune, a set of prompts, and someone else's compute.
This guide covers what those services actually do, what the same result costs if you do it yourself, the specific things that make an AI headshot read as fake, what happens to your uploaded photos, and when you should just book a photographer instead.
Want to run it yourself? Rangy runs the portrait models on your own API key, and your photos stay on your machine.
Try it freeWhat does an AI headshot generator actually do?
It takes the photos you upload, uses them as identity references, and generates new images of you in different outfits, backgrounds and lighting. Some services fine-tune a small model on your face; others feed your photos to an editing model as references. Either way, the output is generated, not photographed.
That last point matters for expectations. Nothing in the output is a photograph of you in that jacket in that room. It is a model's reconstruction of what you might look like there, anchored to your face.
Which is why identity preservation is the whole game. A headshot that looks great but slightly not-like-you is worse than a plain phone photo, because the wrongness is uncanny rather than merely unflattering.
What do AI headshot services cost?
Typically $30–60 for a batch of around 40 images, with faster turnaround or more variety at higher tiers. Running comparable models yourself costs roughly six to nine cents per image, so the same forty images land near $2.40–3.60 in provider fees.
Before treating that as a straight saving, be honest about what the service is doing for the difference:
- Prompt and setting design. Someone chose forty combinations of outfit, background and lighting that generally flatter people. That is real work you now do yourself.
- Culling. Services generate far more than they show you and discard the failures. Doing it yourself means looking at the failures.
- A tuned pipeline. Consistent settings that avoid the common artefacts, arrived at through volume you do not have.
So the honest framing is not "services are a rip-off". It is that you are paying roughly $40 for curation and convenience. If your time is worth more than that for an hour's fiddling, buy the service. If you enjoy the control, or you want to iterate repeatedly, run it yourself.
Can you make AI headshots yourself?
Yes. Feed a good source photo to an identity-preserving editing model, describe the outfit, background and lighting you want, and generate variations. The critical instruction is the one telling the model to keep your face unchanged — without it, results drift toward a generic average.
The workflow is short:
- Start from one good photo. Sharp, front-facing or slight three-quarter, even light, neutral expression, nothing covering your face. This single input matters more than any prompt.
- Change one axis at a time. Outfit, background, lighting. Changing all three at once is where identity drift creeps in.
- Lock the face explicitly. State that the facial features, proportions and expression must stay exactly as in the source.
- Generate several and cull hard. Expect to keep one in four. That ratio is normal and is what the services are absorbing for you.
The instruction that matters most: tell the model what must not change. "Keep her facial features, proportions and expression exactly as in the reference; change only the clothing and background." Editing models drift toward conventional averages unless told otherwise — the same principle covered in the portrait retouching guide.
Which model is best for AI headshots?
Nano Banana Pro, because identity preservation is the constraint that matters and it is the most reliable at leaving a face alone while changing everything around it. Nano Banana 2 works at about a third less cost if small deviations are acceptable.
| Model | Why for headshots | ~Cost at 2K |
|---|---|---|
| Nano Banana Pro | Best identity preservation; follows conservative instructions | $0.09 |
| Nano Banana 2 | Nearly as good, cheaper; occasional small drift | $0.06 |
| Flux Kontext Pro | Very conservative, preserves input structure well | $0.04 |
| Skin Enhancer | Rebuilds texture if a result came out waxy | $0.05 |
Rates checked against Rangy's live pricing tables, August 2026. Rates vary by provider and change over time.
The detailed comparison of the two Nano Banana models is in Nano Banana 2 vs Nano Banana Pro. For headshots specifically, the identity-preservation gap is exactly the case where Pro earns its premium.
What makes an AI headshot look fake?
Four tells, in order of how often they appear: plastic over-smoothed skin, malformed clothing details like collars and lapels, lighting that does not match the claimed setting, and subtle facial drift where the person is nearly but not quite themselves.
- Plastic skin. The most common and most fixable. Ask for blemish removal only, name pores and fine lines as things to preserve, and never use the words smooth or flawless.
- Clothing that does not resolve. Collars that merge into necks, lapels with impossible geometry, buttons that drift. Models are weaker on garment structure than on faces. Zoom in on the collar before you use anything.
- Lighting that contradicts the setting. An office background with studio strobe falloff reads as composited even when everything else is right.
- Facial drift. The hardest to notice on yourself and the most damaging. Put the source and the result side by side at full size before deciding.
A practical check: show a candidate to someone who knows you, without context. If they hesitate, the identity drifted, and no amount of retouching fixes that.
What happens to the photos you upload?
That depends entirely on the service's terms, and it is the least-examined part of this decision. You are uploading a set of images of your face to a third party. Some delete inputs after processing, some retain them, and policies differ on whether uploads may be used to improve models.
This is worth reading the terms for rather than assuming, because the input here is biometric-adjacent and unusually personal. Three questions worth answering before uploading:
- How long are inputs retained, and can you delete them on request?
- Are uploads used for training or product improvement, and is that opt-in or opt-out?
- Where is the processing done, and under which jurisdiction?
Running the models yourself changes the shape of this rather than eliminating it. Your source photo still goes to the model provider's API in order to be processed — that is unavoidable with hosted models. What changes is that there is no intermediate service retaining a library of your face, the provider's terms are the only ones that apply, and the outputs land on your own disk rather than in someone's gallery.
If you want no third party at all, a locally-run open-weight model on your own hardware is the only complete answer, at the cost of setup and a capable GPU.
Are AI headshots acceptable for LinkedIn?
Generally yes, and they are now common enough not to raise eyebrows — provided the image still looks like you. The professional risk is not that it was AI-generated; it is showing up to a meeting looking noticeably different from your profile photo.
The reasonable standard most people apply: the photo should represent how you actually look, presented well. Changing your outfit and background is the digital equivalent of getting dressed up for a shoot. Changing your jawline, age or body is a different thing and will be noticed in person.
Two practical notes. Some platforms and employers have their own policies on AI-generated profile imagery, and those are worth checking in regulated industries. And professional bodies in fields like law and medicine sometimes have guidance on representation in marketing material.
When should you just hire a photographer?
When the photo carries professional weight — executive profiles, press, speaker pages, regulated professions — or when you need full-body, environmental or team shots. Generation is weakest exactly where the stakes are highest, and a real shoot gives you something no model can: a photograph of you.
I build a tool that competes with this, so weigh it accordingly — but four cases genuinely call for a camera.
- The photo is a credential. Executive bios, press kits, conference speaker pages, book jackets. If it will be scrutinised, the small artefacts matter and the provenance question is a distraction you do not need.
- Anything beyond head and shoulders. Hands, full-body posture and how a garment actually hangs remain weak points. Headshots crop out most of the hard problems, which is why they work.
- Team consistency. A wall of headshots shot by one photographer under one lighting setup looks coherent in a way independently generated images do not.
- Regulated or high-trust fields. Where representation rules exist, or where clients would react badly to discovering a photo was generated, the saving is not worth the exposure.
There is also a middle path worth knowing: get one good real photograph taken, then generate outfit and background variations from it. You keep a genuine photograph as the anchor and get the variety cheaply.
Frequently asked questions
Are AI headshots worth it?
For most professional profile use, yes — they are dramatically cheaper than a shoot and good enough that the result does not read as generated, provided identity is preserved. They are not worth it where the photo carries real professional weight, such as executive bios or press material, or where you need anything beyond head and shoulders.
How much do AI headshots cost?
Services typically charge $30 to $60 for a batch of around forty images. Running comparable models yourself costs roughly six to nine cents per image, so forty images is about $2.40 to $3.60 in provider fees. The difference is largely curation and convenience rather than compute.
Can you make AI headshots without a subscription?
Yes. Editing models that preserve identity can be run directly on a pay-per-image basis with your own provider account, so there is no monthly fee and nothing is charged in months you do not use it. You supply the source photo and the prompts, which is the work the service was doing for you.
Do AI headshot services keep your photos?
It varies by service and is set out in their terms rather than being standard across the category. The questions worth answering before uploading are how long inputs are retained, whether they can be used for training or product improvement, and where processing happens. Running models yourself avoids an intermediate service holding a library of your face, though the source image still reaches the model provider's API.
Why does my AI headshot not look like me?
Usually because the instruction did not explicitly lock the face. Editing models drift toward conventional averages unless told to preserve facial features, proportions and expression exactly as in the source. Changing outfit, background and lighting all at once makes drift more likely; change one axis at a time and compare against the original at full size.
Is it okay to use an AI headshot on LinkedIn?
Generally yes, and it is common enough not to be remarkable, as long as the image still represents how you actually look. The practical risk is not the technology but the mismatch — arriving at a meeting looking noticeably different from your profile photo. Some employers and regulated professions have their own policies worth checking.
What kind of photo should I start from?
One sharp, well-lit photograph, front-facing or a slight three-quarter angle, with a neutral expression and nothing obscuring your face. Even, soft light beats dramatic light as a starting point because it gives the model clean information to work from. The quality of this single input affects the result more than any prompt wording.
The bottom line
Buy the service if you want forty curated options for $40 and an hour of your time back. Run it yourself if you want control, repeat iterations, or your photos not sitting in a third party's library. The models are the same; you are choosing who does the culling.
Whichever route you take, the quality bar is identity, not polish. A slightly plain headshot that unmistakably looks like you beats a beautifully lit one that does not, because the second kind gets noticed in the worst possible way — in person, by someone who was expecting somebody else.
Check the collar, check the skin texture, and put the result next to your source photo at full size before you use it anywhere.
"This software has increased my workflow speed tenfold, and the output quality it has delivered in my work has been exceptional."
Run the portrait models yourself
Rangy runs Nano Banana Pro, Nano Banana 2 and Freepik's Skin Enhancer on your own API key, from about six cents an image — with every file saved to your own disk.
Download Rangy Free Or read the retouching guide first →Mac & Windows · Free plan, no credit card · 5 generations a day
The model rates come from Rangy's live pricing tables, checked on 14 August 2026, and will drift as providers change them. The $30–60 service figure is a range across the category rather than a quote from any one provider — pricing and batch sizes differ and change, so check the service you are considering. I have deliberately not named specific services or characterised their data-retention policies, because those differ, they change, and getting one wrong would be unfair to the company and misleading to you. Read the terms of whichever you use. Illustrations were generated with GPT Image 2 at 2K.
I develop Rangy, which competes with these services, so this is not neutral. The cases for buying the service instead are in what the services cost, and the cases for hiring a photographer are in when to hire a photographer.