What thousands of generated headshots taught us about how people want to be seen

Team GoStudioJanuary 13, 20268 min read

Last updated: January 13, 2026

There is a gap between the photo people say they want and the photo they actually keep.

Contents

The finding

Across [N] headshots generated on GoStudio between [start date] and [end date], the pack people opened most often was [pack], ahead of [runner-up] by [margin]. The images they kept were not the ones I would have predicted from the packs they chose. [One sentence stating the finding plainly.]

What we looked at

Everything below comes from aggregate usage data: which pack someone opened, how many images that produced, and which of those images they downloaded. No individual accounts, no customer images, and nothing that could identify a person. I want to be specific about that before I show you a single number.

A generation counts once per image produced. A download counts once per image saved, so saving the same image twice does not count twice. I excluded internal accounts, my own test runs, and any pack with too little traffic to say anything honest about.

That leaves [N] generations and [N] downloads across [N] packs, between [start date] and [end date]. Where a cut below covers less than that, I have said so next to it rather than at the bottom of the page.

What people pick

Somebody opening GoStudio is choosing from more than two hundred packs. I assumed that much choice would spread people out, with a few popular options and a long tail of everything else. That is not what happens.

[The distribution, stated plainly: which subcategories take the largest share, and how concentrated the top few are against the tail.]

[Chart: pack selection by subcategory, built from the real numbers. Label the sample size and the date range on the chart itself so it survives being screenshotted.]

The part I did not expect was the split between packs built for different subjects. [What the supported_classes cut shows, described as what the numbers say rather than as a claim about people.]

The gap between generated and kept

People generate broadly and keep narrowly. The pack someone chooses is aspirational. The image they download is the one they can actually imagine other people seeing.

This is the number I think about most. Generating an image and keeping one are two different decisions, and the distance between them is the closest thing I have to an honest signal. Almost nobody writes in to say a photo is wrong. They just do not download it.

[Generated against downloaded, per pack. Give the overall ratio first, then the packs at either end of it.]

[The packs where the ratio is worst, and what those packs have in common. Cross-reference the regeneration rate here: a pack people retry repeatedly is a pack where likeness is failing, and that is a product problem rather than a taste one.]

The reading I keep coming back to is that people generate broadly and keep narrowly. The pack somebody picks is aspirational. The image they download is the one they can actually picture other people seeing. [Test this against the numbers before it goes out. If they do not support it, cut this paragraph rather than softening it into something that survives either way.]

It differs by market

GoStudio prices by country, which means I can see roughly where people are when they choose a pack. I went in expecting the difference between markets to be small. It is not.

[The geographic split: which packs over-index in which markets, and by how much.]

[Say what the numbers show and stop there. What a photo has to look like to be taken seriously is set by whoever is on the other side of the application, and that expectation is local. Describing the pattern is fair. Explaining a culture from download counts is not.]

What we changed because of this

Publishing numbers is easy. Changing something because of them is the part that costs, so here is what actually moved.

[Which packs were built, retired or reordered as a result, and what the pack order looks like now. If the honest answer is that nothing changed, say that instead. It is a more interesting sentence than a manufactured one, and a reader can tell the difference.]

What we still do not know

None of this shows why somebody chose something. It shows what they clicked. A person who downloads one image out of forty might have found the one that looked like them, or might have given up on the other thirty nine. The numbers read exactly the same either way.

Anyone who tells you their analytics reveal intent is guessing. I can see the choice and not the reason behind it, and I would rather say that here than let you assume otherwise.