You have been using an app for a fortnight on its middle tier, and the pricing page has one row you cannot account for. The top plan costs a good deal more than the one you are on, and the only line it adds that you have any interest in is an AI girlfriend app video call. It is worth knowing what that row sells before you pay for it, because in most apps a video call is not video. Nobody is filming anything. The thing moving on your screen is assembled out of your own voice, a beat after you stop talking.
What an AI girlfriend app video call actually is
Four systems run in a row to produce one. Your speech is recorded and transcribed. The transcript goes to the same language model that writes your chats. The reply is read out by a synthetic voice. Then a fourth system animates a picture of the character so the mouth moves in time with that audio. The character is not being filmed; it is being drawn to fit a sound generated a fraction of a second earlier.
Which means the call adds nothing to the conversation. It is the voice feature with a picture bolted on: same character, same personality settings, usually the same memory. What varies is how that picture gets made, and there are broadly three ways. Some generate an animated portrait as the audio plays. Some drive a rigged three-dimensional model, the way a game animates a character. Some play back a handful of clips recorded in advance and loop them behind the voice. Apps rarely say which they sell, and the marketing word for all three is the same.
Generated, or played back
If you get one answer out of an app before you upgrade, make it this one: those two ends of the range cost wildly different amounts to run and are routinely sold at the same price.
You can usually tell inside one call. Ask for something that would have to change the picture — a coat, a different setting, standing rather than sitting. A system drawing frames as it goes will attempt it, well or badly; a playback system carries on as it was and answers in words only. Then stop talking and watch the silence. A clip library keeps moving to its own schedule regardless of you. Then watch for a repeat. The same gesture twice inside five minutes means a short library on a loop.
None of which makes playback worthless. A well-cut loop is less distracting than a shakily generated face. The point is knowing which of the two you bought at a top-tier price.
Why a face is harder than a voice
Audio is forgiving. A pause before a reply reads as someone thinking, and a slightly synthetic voice stops being noticeable after a few minutes. A face has no such cover: the same delay looks like a dropped connection, and a mouth landing a quarter-beat behind the words is visible every time.
Wikipedia's summary of the uncanny valley puts the mechanism in one line: "If an animated character looks more human than its movement, this gives a negative impression." That describes a photorealistic portrait driven by imperfect lip-sync exactly, and it explains something counterintuitive: a deliberately stylised avatar often holds up better across ten minutes than the glossier option, because it is not promising motion it cannot deliver.
A voice can absorb a pause. A face cannot, and the better it looks the less it can.
The length of that pause decides whether the audio version works too, and the two-day test in our guide to AI girlfriend voice chat measures it on a free tier, before video is a question at all. One more check: whether the face in the call is the face in your pictures. The animation and image systems are separate, so they often disagree — the same problem that turns up in generated photos.
Whether your camera is part of it
The word "call" implies two cameras; in most of these apps there is one stream running one way, and nothing of you goes back but sound. Some apps do ask for the camera, usually so the character can remark on what it sees. That is a different feature under the same name, and it brings its own questions — how long frames are kept, whether they are used to improve the service, whether deleting the conversation deletes them. Decline the prompt and the call still works in most apps.
Why video sits at the top of the price list
Every second of a call spends speech recognition, model tokens, speech synthesis and frame rendering at once. Four meters running together makes it the most expensive thing these apps do per minute, and the thing most reliably kept for the top plan.
There is a second reason that has nothing to do with cost. A top tier needs something the tier below cannot have, or nobody moves up. Video suits that job because it is the one feature you cannot evaluate from a screenshot or a store page — it has to be experienced, and experiencing it means paying first. Our guide to which tier is the real product covers the rest of that structure.
So it is usually metered even inside the tier that includes it: a bundle of call minutes a month, or a credit balance spent per minute. Three questions answer it from the pricing page. How much video the tier includes, in units you can picture. What happens when it runs out — the call stops, drops to audio, or asks you to buy more. And whether the balance is shared with image generation, because if it is, a week of pictures quietly costs you your calls.
How to test it when the free tier gives you nothing
Video is the feature least likely to come with a free allowance, so the usual free-plan test does not apply. Two things work instead.
First, ask support whether the avatar is rendered live or played from clips. How readily the answer arrives tells you something by itself.
Second, if you are going to pay, buy exactly one month of the top tier — and check first that the way back exists. Google Play's guidance on changing a subscription says moving between plans works only if the app's developer allows it, that a change may cost the full price of the new plan rather than the difference, and that remaining paid time is the developer's business rather than the store's.
Then use the month deliberately. Make one ten-minute call in the first week and one in the fourth, and notice who started each: you, or a notification. If both were prompted rather than chosen, you have your answer about renewal. If you would rather test something than build a shortlist first, the app we currently recommend is a fair place to begin — run the checks rather than take our word for it.
Now the part that cuts against the upgrade. The real argument for voice was that it got the app off a screen. Video undoes that: it wants a screen, your eyes, and a stretch of time in which you are doing nothing else — the narrowest slot in most people's week. If your real use of one of these apps is fifteen minutes before sleep, video is the feature least likely to earn the step up, and the same money goes further on whichever tier buys more memory.
Questions people ask
Is an AI girlfriend app video call a real video call?
No. There is no camera at the other end and nothing is being filmed. A picture of the character is animated to match a synthetic voice, either generated as the call runs or played from clips made in advance.
Do I need to turn my camera on?
In most apps, no — the stream runs one way and only your audio is sent. A few ask for the camera so the character can respond to what it sees, which is a separate decision with its own privacy questions.
Is video included in any free plan?
Rarely, and rarely for more than a minute or two. It is the most expensive feature these apps run, so it tends to be kept for the highest tier and metered even there — which makes one short paid month the only honest way to try it.
Disclosure. MoonSyrup may earn a commission if you sign up through a Visit link on this page, at no cost to you. It does not change what we write. How we earn.