The default way people trial a headshot generator is to feed it 15 flattering photos of a cooperative face and admire the output. It is the cheapest possible test and it measures nothing.
Every headshot product handles the easy case. That is table stakes. The spread between tools is entirely in the hard case, so an easy-input trial returns a tie and you pick on price or design.
Pick the input to break it
Hard means, roughly: glasses with a strong prescription, tightly curled or very voluminous hair, darker skin tones, distinctive facial hair, visible scars or birthmarks, and any pronounced asymmetry. Which is to say, most real faces.
If your own face is one of those, use it: you will judge the portrait better than any metric. You will judge likeness far better than any similarity metric, and likeness is the thing that actually varies.
Then ask one question of the results: does it still look like the person. Photographic quality is the easy half and the half everyone measures.
The input mistake that gets blamed on the model
Uploading 15 photos from a single afternoon. Same shirt, same wall, same light.
The model has no way to know which features are you and which are Tuesday. It learns the shirt as part of your identity and the outputs come back oddly attached to it. Then people conclude the tool is bad.
Vary days, clothing, angle and light. It feels like it should make things less consistent and it does the opposite.
Do it on the free tier
Both tests run fine without paying, which is what free tiers are for. Ours are at betterpic.io/free-tools, and the honest framing is that they exist so you can find out whether this works on your face before spending anything. If it does not work, you have lost 10 minutes and learned something specific.
While you are evaluating, check the incentives
This category has an unusually heavy affiliate layer. A large share of the "best AI headshot generator" listicles you will find are commission-driven, which does not automatically make them wrong and does explain why the rankings are so inconsistent between them.
We run an affiliate programme ourselves, so this is not a stone I can throw from a great height. The practical advice is to check whether a comparison discloses, and to weight tests you ran yourself far above any of them.
Our own writing is at the blog, and the Elucient case study is a reasonable example of the level of detail worth expecting from anyone's customer story.
What good looks like
You have run one hard face through two tools on the same day with varied inputs, and had someone who knows that face pick the real photos out of a mix. That is a 20 minute experiment and it beats every review you will read, including the ones written by people with no commission at stake.
More from this series
- Crop ratios and framing conventions across 5 professional photo genres, as a reference table
- Colour management for portfolio sites, and the 4 professions where getting it wrong loses the client
- The same portrait at 5 sizes: what breaks at 1200px, 400px, 96px, 48px and 15mm of print
- Photographier une entreprise entière est un problème d'agenda déguisé en problème de photographie
On BetterPic
Top comments (0)