"Can you do that AI thing too?" — that question now shows up in almost every wedding and corporate-event inquiry. If you answer "yes," you should know exactly what the customer means. Because "AI photo booth" actually covers three completely different things, which cost different amounts and go wrong at very different rates.
Three things, one word
Background removal without a greenscreen. A model separates the person from the background without a green cloth hanging anywhere in the room. It saves on setup and is the least noticeable of the three cases: the result looks like a normal photo in front of a chosen backdrop. The computation happens locally on the event PC — there are no ongoing per-image costs.
Style transformation. The guest becomes a cartoon character, a superhero, an anime figure. This is what most people actually mean when they say "AI" — and the only one of the three that genuinely costs money per image, because a large image model has to compute it. That computation doesn't run on the event PC, it runs at a provider online.
Upscaling and touch-up. Sharpening, noise removal, skin smoothing. Useful, but not something guests talk about at the party — and usually already handled in the camera or the layout.
Anyone who writes "AI" in a quote should spell out which of the three they mean. A customer expecting superheroes who gets greenscreen background removal is disappointed — and, the other way round, someone ends up paying for compute time they never actually needed.
What a transformation really costs
Let's run the numbers for a typical wedding: 120 guests, of whom experience shows 40 to 60 percent use the booth, many more than once. Realistically that's 80 to 120 sessions over the evening. If every second session triggers a style transformation, you land at 40 to 60 transformations per event.
At BoothDock, a credit package with 100 transformations costs €14.90 a month, 300 cost €39.90, 750 cost €89.90. Depending on the package, that works out to between 12 and 15 cents per transformation. For our example wedding: roughly €6 to €9 in material cost for the whole evening — less than a roll of print paper.
Important for your calculations: credits apply to the whole account, not per box. If you run three events in one weekend, they draw from the same pool. And one credit covers a transformation plus one retry — the guest gets to change their mind once without costing you a second time. That's not a minor detail: without this rule, every unhappy guest becomes a cost factor, and your staff start rationing retries.
The wait time is your real risk
A large image model needs seconds, not milliseconds. Our own measurements on September 8, 2026 came out at 22 to 33 seconds per transformation, depending on style and load. At the booth itself, that feels long — especially with a line forming behind the guest.
Two things decide whether that becomes a problem. First: does the guest get their original photo immediately, or do they wait for the one result? If both the original and the AI version land in the gallery, the wait feels relaxed — printing can already be underway. Second: what happens when the service doesn't respond? A booth that just stops is a failure at a party. A booth that quietly falls back to the original photo and keeps going is a booth that works.
Ask every vendor exactly that question: "What does the booth do when your AI is down?" The answer separates built software from demoware.
Where the photo goes — and how long it stays there
For the transformation, the guest's photo leaves the event PC. That makes it data processing like any other, and the usual questions apply: which provider does the computing? In which country? Is there a data processing agreement? And above all: is the input photo deleted afterward — or does it stay around "for quality improvement"?
You should get that last question answered in writing. At a corporate event with a works council, it's the first one asked. At BoothDock, the transformation runs at a European provider, and the input photo is deleted immediately after the transformation.
Since August 2026: labelling is required
Under the transparency obligations of the EU AI Act (Article 50), AI-altered image content must be labelled. For photo booth operators that means two things in practice.
Machine-readable: the image file itself needs a marker — metadata that says: this image was generated by a model. The guest doesn't see it, but platforms and tools can read it out. It costs nothing, doesn't change the photo, and should be a given. Ask your vendor whether they set it; most still don't, to this day.
Visible is the second layer: a notice a human actually notices. The EU Commission has provided its own symbols for this since August 2026 — using them is explicitly voluntary, the labelling obligation itself is not. Where the notice sits is your call: a mark in the image, a line in the gallery, a sign at the booth. What matters is that it exists, and that it's in the guest's language.
How to spot a serious vendor
- Curated styles instead of a random generator. A model that distorts faces isn't fun at the party — it's a complaint. Good vendors test their styles individually.
- The original is preserved. A vendor that only delivers the AI image takes away the guest's choice — and takes away the photo the customer will actually want later.
- An outage falls back to the original photo, not an error message.
- Clear cost per transformation, not "contact us."
- Labelling built in, without you having to ask.
- Deletion of the input photo, promised in writing.
In short
Style transformation is the first AI feature guests actively line up for. It's cheaper than most people think — one evening costs less than a roll of print paper. It only gets expensive if the wait time blocks your flow, if you have to fix your privacy story after the fact, or if the labelling is missing and a corporate client asks about it.
Sort out these three points before your first event and you have a selling point. Sort them out afterward and you have a problem.