Questions to Ask Before Buying Any AI Headshot Tool (Not Just Ours)

Last updated: July 15, 2026.

A note on how this page was built, upfront: we set out to write this as a roundup of real Reddit threads recommending AI headshot tools. We ran into an honest limitation, Reddit blocks most automated access to its threads, and the summaries of "what Reddit says" that are searchable elsewhere are secondhand paraphrases we couldn't verify against the original discussions. Rather than dress up unverifiable secondhand summaries as if we'd read the threads ourselves, we rebuilt this as something we could actually stand behind: the specific questions worth asking before you buy from any AI headshot tool, ours included, based on the patterns we've seen across this entire category while researching the rest of this site.

The short answer

• Ask what happens to your photo after generation: is it deleted, and on what schedule, or kept indefinitely.
• Ask whether the price you see is the price you pay, or whether the usable output requires an upsell.
• Ask how many source photos are actually required, and whether more photos genuinely improve the result or just add friction.
• Ask whether there's a free sample before you pay, and whether that sample is representative of the paid output or deliberately worse.
• Ask what the refund policy actually covers, and whether it's time-limited, usage-limited, or both.
• Ask whether the tool trains on your photos or keeps a face embedding after your session ends.

What happens to your photo, specifically

This is the question most people don't ask and should. "We take privacy seriously" on a pricing page means nothing on its own. The specific, answerable questions are: is the photo processed in memory or written to disk, is it deleted automatically and on what timeline, is it ever used to train or fine-tune a model, and is a face embedding (a mathematical representation of your face, distinct from the photo itself) retained after your session. A tool that can't answer these plainly, in writing, on its own site, is a tool you should hesitate on regardless of how good the sample gallery looks.

On our own handling: photos are processed in memory, never used to train a model, and deleted on a fixed schedule rather than kept indefinitely. Full detail is on our security page. We're not claiming this makes us unique, several other tools in this category have similarly good policies, we're pointing at it because it's exactly the kind of specific, checkable claim you should be looking for from whichever tool you actually pick.

Is the advertised price the real price

A pattern worth watching for across this category: a low headline price ("$9," "free") that turns out to gate the actually-usable output behind an upsell, watermark removal, higher resolution, or additional styles that weren't clear from the pricing page. This isn't necessarily dishonest, watermarked free samples are a completely reasonable way to let you test quality before paying, but it's worth reading the fine print on exactly what a given price tier includes before assuming the number on the homepage is what you'll actually spend.

The practical check: before you pay anything, find the specific list of what's included at that price (photo count, resolution, watermark status, style count, commercial usage rights) rather than going off the headline number alone.

How many photos does it actually need, and does more help

This varies genuinely by underlying technology, not just by tool preference, and it's worth understanding why. Tools built on per-user fine-tuning (training a small custom model on your specific photos) generally do benefit from more source photos, more angles and expressions genuinely improve the trained model's likeness. Tools built on single-photo, no-training approaches (identity-preserving adapters or multimodal models reading one reference image directly) don't meaningfully improve with additional photos beyond the first one or two, since there's no training step for more data to improve.

If a single-photo tool is asking you to upload ten photos anyway, that's worth a moment's skepticism about why. If a tool that requires extensive training only asks for one or two photos, the output quality may suffer for it. Neither approach is universally better, but the number of required photos should roughly match the underlying technology's actual needs, not just be a fixed number for its own sake.

Is the free sample honest, or deliberately worse

A free tier exists for one of two reasons: as genuine loss-leader marketing (the free result is representative, just limited in volume or resolution) or as a deliberately hobbled teaser designed to make the paid tier look better by comparison than it would look on its own merits. There's no perfect way to tell which one you're looking at from outside, but a reasonable check is whether the free sample uses the same underlying model as the paid tier (some tools genuinely downgrade to a worse or older model for free users) or just the same model with a lower resolution and a watermark.

Our own free tier uses the exact same model as every paid generation, the difference is volume (3 a day versus a larger one-time pack), resolution, and the watermark, not a downgraded model. That's a specific claim you can hold any tool to, including us: ask directly whether the free tier runs the same model as paid, or a cheaper substitute.

What does the refund policy actually cover

Read the specific conditions, not just whether a refund policy exists. Common real-world limits: a time window (24 hours, 7 days) after which no refund is possible regardless of usage, a usage cap (refund voided if you've downloaded or generated beyond a small number of photos), or both stacked together. Neither of these is unreasonable on its own, a business needs some protection against someone generating 100 photos and then requesting a full refund, but you should know the actual numbers before you pay, not discover them after a bad experience.

Does the tool remove things that are part of your identity

This is a real, specifically documented failure mode, not a hypothetical one. A UC Berkeley Law researcher's empirical test of 25 widely used paid AI headshot platforms found that 22 of the 25 removed a hijab entirely from photos of hijab-wearing women and replaced it with AI-generated hair, automatically, with no option to consent to or prevent the removal. The same category of problem applies to accessibility devices (wheelchairs, hearing aids, prosthetics) and other visible identity markers that a tool wasn't explicitly built to preserve.

If this applies to you, it's a specific, testable thing to check with a free sample before you pay, not something to just assume a tool handles correctly because it seems unlikely a reputable-looking product would get it wrong. The Berkeley research found this failure across the large majority of tools tested, including several well-known, widely used ones, so "the tool looks polished" is not a reliable signal that this particular problem has been addressed.

How to actually judge a sample gallery, not just admire it

Every tool's homepage shows its best output. That's normal marketing, not deception, but it means a sample gallery tells you almost nothing about your own likely result unless you know what to look for. A few specific things worth checking instead of just scrolling past:

Look for variation within a single person's set, not just variety across different people. A gallery of ten different attractive people each shown once tells you the tool can occasionally produce a great result. A gallery showing the same person across five or six generations tells you something closer to what you'll actually get: the realistic range, including the misses, not just the single best roll.

Look at hands, ears, and background edges specifically, the classic weak points for image generation models generally. A sample gallery that only shows tight face crops with the edges conveniently cut off is hiding exactly the areas most likely to show artifacts.

Look for stated turnaround time, not just "fast." Current single-photo, no-training tools typically return results in under two minutes; anything advertised in "hours" is very likely still using an older per-user-training approach, which isn't necessarily worse, but you should know which one you're signing up for before you upload anything, not discover it while waiting.

A genuinely fair note on where we might not be the right fit

In the interest of the same honesty we're asking of every tool in this category: if you need a large team package with per-seat admin controls, or a tool specifically built around event photobooth logistics, there are competitors that specialize in exactly that in ways we currently don't. We cover honest, specific comparisons against several of them in our comparison pages. The questions above apply whether you end up choosing us or someone else, that's the actual point of this page.

FAQ

What's the single most important question to ask before buying an AI headshot tool?

What happens to your photo after generation, specifically whether it's deleted, on what schedule, and whether a face embedding is retained. Most pricing pages don't volunteer this detail, so it's worth asking directly if it isn't already clearly stated.

Do more source photos always produce a better AI headshot?

Not universally, it depends on the underlying technology. Tools that fine-tune a custom model per user do benefit from more photos. Tools using single-photo, no-training approaches don't meaningfully improve with additional photos beyond the first one or two.

How can I tell if a free AI headshot sample is representative of the paid quality?

Ask directly whether the free tier uses the same underlying model as the paid tier, with the difference being volume, resolution, and watermark, or whether it's a cheaper, different model used specifically to make the paid tier look better by comparison.

What should I check in a refund policy before paying?

The specific time window (often 24 hours to 7 days) and any usage cap (refund voided past a certain number of downloads or generations), not just whether a refund policy exists at all. Both details matter more than the headline "refunds available."

Do AI headshot tools remove things like hijabs or accessibility devices?

Often, yes. A UC Berkeley Law researcher's empirical test of 25 paid AI headshot platforms found 22 removed a hijab entirely and replaced it with AI-generated hair, with no consent option. It's a reasonable, specific thing to test with a free sample before trusting a tool with a paid purchase if it applies to you.

How do I judge a tool's sample gallery if every homepage shows its best work?

Look for the same person shown across multiple generations (the realistic range, not just one great roll), check hands and background edges specifically rather than tight face crops, and note the stated turnaround time, since that tells you which underlying technology you're actually signing up for. Covered in more detail above.