How I test AI headshot generators

Ten tests, the same for every generator, each graded from 1 to 5 on a fixed scale. The overall grade is their average plus my verdict, which can move it by up to one point. Every test is run by hand on an account I use, and the photos, counts, panel results and correspondence behind each grade go in the full test results published with the review.

Written by Miles Foster, Owner and tester, RankDude.

The RankDude mascot holding a laptop with a check on its screen

Each generator gets the same two sets of selfies: 15 of me and 15 of a second adult with a different skin tone, who agreed in writing, all taken on one phone in mixed light, plus three deliberately bad photos to see what gets flagged. I ask for the same three styles every time (a grey studio backdrop, an office, and outdoors) on the package closest to 100 headshots. Nothing is graded from a sample gallery or a press kit.

Each review states how the account was obtained: bought at the public price, used on a free tier, or provided by the vendor. A vendor-provided account never changes a grade, and a vendor never sees a review before it is published.

How I rate

Every tool gets one score out of 5.

First, I run the same 10 tests on every tool and give each one a grade from 1 to 5. The average of the graded tests is the starting point; a test that cannot be graded is left out of it, and the review says how many tests the score is based on.

Then I add my verdict: what the tests cannot measure, such as how the tool holds up after the test weeks, how billing and cancelling work in practice, and how it compares with the others in the category. My verdict can move the score up or down by one point at most.

Every review shows how much my verdict changed the score, and why, in its scores table.

What a grade means

5	Excellent
4	Good
3	Average
2	Below average
1	Poor

The final score is the average of the graded tests (ten when every test can be graded), to one decimal place, plus my verdict, which can move it by up to one point. If a test couldn't be finished properly it gets no grade, and the review says how many tests the score is based on. When a free tier delivers only a few headshots, the counted tests use what it delivers; I test them again on a paid package when I can.

The 10 tests

1.Likeness

For each subject, 20 headshots picked at random are mixed with 5 real photos and shown to a panel of five people who know that person. Each answers whether the picture shows them. The share of AI headshots recognised, across both subjects and all panellists, is the grade; the real photos check that the panel is paying attention.

5: 90% or more recognised
4: 75% to 89%
3: 60% to 74%
2: 40% to 59%
1: under 40%

2.Realism & Artefacts

40 headshots per subject, picked at random, are inspected at 100% for faults in hands, teeth, ears, eyes, hairline, glasses, jewellery, collars and buttons, skin texture and background text. A headshot is clean when it has none. The share of clean headshots out of 80 is the grade, and each kind of fault is tallied.

5: 90% or more clean
4: 75% to 89%
3: 60% to 74%
2: 40% to 59%
1: under 40%

3.Style Variety

I check whether each of the three requested styles was delivered as asked, then count distinct combinations of background, outfit, pose and lighting in the set. Near-identical shots with a small crop or colour change count once.

5: all three styles honoured, 10 or more distinct looks
4: all three honoured, 6 to 9 looks
3: two styles honoured, or 3 to 5 looks
2: one style honoured, or near-identical shots
1: requested styles ignored

4.Usable Shots per Batch

Every headshot delivered for both subjects is sorted. A usable headshot is recognisable, has no artefacts at 100%, has a sharp face, natural skin and a crop that works as a profile photo. The share and the number of usable shots are both recorded; the number feeds the pricing test.

5: 40% or more usable, at least 20
4: 25% to 39%, or 40% or more but fewer than 20
3: 15% to 24%
2: 5% to 14%
1: under 5%

5.Turnaround Time

The clock runs from confirming the order with the selfies uploaded until the full set can be downloaded, for both subjects. The longer of the two is the grade; the promised delivery time is reported alongside.

5: under 30 minutes
4: 30 to 59 minutes
3: 1 to 2.9 hours
2: 3 to 24 hours
1: over 24 hours, or not delivered

6.Inputs & Ease of Use

I record the minimum number of selfies required and the guidance given, then upload the set with three bad photos mixed in (sunglasses, a group shot, a blurred face). I note which are flagged before the model is trained, and time the steps from sign-up to a placed order.

5: 12 or fewer selfies, all three bad photos flagged, under 10 minutes
4: two or three flagged, under 20 selfies required, under 20 minutes
3: one flagged, or 20 or more selfies required, or 20 minutes or longer to order
2: none flagged
1: the order could not be placed

7.Edits & Retakes

Each option is tried on a delivered headshot: a free retake or regeneration, a background change, an outfit change, a crop to new proportions, an expression or pose variation, and a human or manual touch-up where offered. Each counts when the result keeps the likeness and looks natural.

5: all six work
4: four or five
3: two or three
2: one
1: none, or no edits after delivery

8.Privacy & Deletion

I read the stated retention period for selfies, the trained model and the outputs, and whether uploads are used to train anything without opting in. Then I request deletion through the stated channel, record the time to written confirmation, and after the stated window check whether the account gallery and the direct image addresses I saved still load.

5: retention 30 days or less, no training by default, deletion confirmed within 7 days and images gone
4: deletion honoured within 30 days, one policy weakness
3: deletion honoured, but retention long or unstated
2: deletion over 30 days, or uploads used for training by default
1: deletion not honoured, or no way to ask

9.Resolution & Usage Rights

The long edge of every downloaded file is measured in pixels, and each is checked for a watermark or visible credit. The terms are read for commercial use (company websites, print, advertising), who owns the images, and any limit on where they can be used.

5: 3,000 pixels or more, no watermark, full commercial rights and ownership
4: 2,000 to 2,999 pixels, full rights
3: 1,024 to 1,999 pixels, or rights limited to profile use
2: under 1,024 pixels, or a watermark on paid packages
1: rights unclear or kept by the vendor

10.Pricing & Value

Every package and add-on is read from the checkout, not the marketing page. The price of the package tested is divided by the usable headshots it delivered, across both subjects. Refund terms, team prices and payment methods are listed as seen at checkout.

5: under $1 per usable headshot
4: $1 to $2.49
3: $2.50 to $4.99
2: $5 to $9.99
1: $10 or more, or hidden costs

Documented but not graded

Team ordering with matching styles, background colour matching for a company, bulk delivery links, mobile apps and full-body or creative portrait styles are checked and described in every review. When they work well they count in the written assessment, but they never add points to the overall grade.

Rules that apply to every test

Real use of the product on a real account, never promotional descriptions or sample galleries.
Identical selfies, styles, tests, thresholds and sample sizes for every generator in the category.
Authorised accounts only; no password sharing, no bypassing of limits or bot checks, and only faces of people who agreed in writing.
Every grade has evidence: the photos, a count, a panel result, a screenshot or correspondence, in the full test results.
A grade changes only when a retest produces new evidence; the review records the date and the reason.

Questions about a specific grade go to the contact page. Makers can submit an AI headshot generator for testing. No AI headshot generator has been through these tests yet; every one I follow is listed on the AI headshot generators page, and reviews appear there as tests finish.