Comparison · 7-min read

I paid for 7 AI headshot apps with the same selfie. Only 2 looked like me.

By HiForrest · Published June 5, 2026

A weekend, $214, seven apps, one face. Three takeaways that will save you both the money and the regret.

I spent a weekend signing up for seven AI headshot apps with the same selfie. I'm going to keep the names blurry here — that's the App Store-safe thing to do and also the polite thing — but I'll tell you what I saw. By the end, I'd spent $214 in upgrade fees and gotten 162 images. Two of the seven apps produced something I would actually use on LinkedIn.

The setup

Same input selfie on every app. Same prompt category (LinkedIn-style professional headshot). I evaluated three things, in this order:

1. Did the face actually look like me? Eye spacing, hairline, jaw shape, skin texture. 2. Was the styling believable for the role? Did the lighting match the supposed environment? Was the wardrobe age-appropriate? Did the background look like a place that exists? 3. What did the experience cost in time and dollars?

Same person. Four prompts. One selfie. ArcFace likeness 0.913 — measured, published, reproducible.

What I saw

The most expensive app produced the least convincing face. That was the first surprise. The second surprise was that the apps requiring 10+ photos didn't outperform the apps requiring one. Throwing more data at the model is not what's failing.

Most of these apps did not lose to us on quality. They lost to us on identity. The face in the picture was not my face.

What's going wrong in the industry

Every AI headshot startup has access to roughly the same base image model. The differentiation isn't in the rendering. It's in what they do with your face before, during, and after generation. Most teams optimize for the demo — magazine-cover skin, perfect teeth, dreamy bokeh. The model produces a great-looking person who is not the person you asked for.

The two apps that worked, including ours, do something subtly different: we score the output against the input on identity preservation and reject anything that drifts. The output looks slightly less magazine-y and considerably more like you.

I'm being deliberately vague about how — there's a copycat economy in this space — but the public-facing claim is the 0.913 ArcFace number, and that number is reproducible. You can install the app, generate a preview for $2.99, and run the comparison yourself.

Pay $2.99 — see your preview

Credit applied to any upgrade. No free-tier tease, no watermark.

The three takeaways

1. Don't pay before you see the result. Five of the seven apps charged the upgrade fee before showing me a single preview. Two showed previews first. Guess which two are the ones I'd recommend. 2. More input photos is not better. The apps that demanded 10–20 reference photos produced the most generic faces. The ones that worked from a single selfie produced the most accurate. 3. Optimize for identity, not aesthetics. A photo that looks like a magazine cover but isn't you is unusable on LinkedIn. A photo that looks like a polished version of the actual you is the entire point.

What I do now

I pay $2.99 to preview, before I pay $35 for the pack. If the previews look like me, I upgrade. If they look like someone else's face wearing my outfit, I don't. The $2.99 is credited toward the upgrade either way, which makes the experiment effectively free if I keep going.