There is a camera icon on the feature list now, beside the microphone, and the tier it sits on costs more. An AI girlfriend app video call is the newest thing in this category and the least explained: the screenshot shows a face mid-sentence, the pricing page gives you a number of minutes, and nothing says what the picture is or why it is sold by the minute when your messages are not. Both answers are the same answer, and it decides whether the tier is worth it.
What an AI girlfriend app video call actually is
It is not a call in the sense your phone means. Nothing is transmitted from anywhere, and the character is not somewhere, looking at a lens. What runs is the voice pipeline with one stage on the end: your speech becomes text, the model writes a reply, a speech system turns it into audio — the three jobs our piece on what voice chat is really doing walks through — and then a fourth system draws a face that appears to be saying it.
The order matters, and you will notice it within a minute. The face is downstream of the words, so it cannot react to you or pull a look while you are still talking. It has nothing to react with: its only input is the audio of a reply that has already been written.
Four stages also take longer than three, so the pause that voice chat already has is longer here — and more conspicuous, because you spend it watching a face do nothing. Interrupting fails for the same reason it does on voice: by the time the mouth moves, the sentence is already on its way to you.
The picture is either a puppet or a fresh frame
Apps build that fourth stage in one of two ways, and the difference explains most of what you will like or dislike.
The first is a puppet. The app holds one prepared likeness, a 2D rig or a 3D model, and animates it: the audio is analysed for which mouth shape belongs where, the lip sync is driven from that, and a small library of blinks and head tilts cycles underneath. It is cheap and perfectly consistent — identical on Tuesday and on Friday. It also loops, which is the thing you eventually cannot unsee: the same tilt at the same interval, the same return to the same resting pose.
The second is to generate the frames outright, with a video model drawing the picture continuously rather than moving a prepared one. Nothing loops, because nothing is prepared, and the motion is looser. So is the face: features drift between frames, and the character is a close relative of yesterday's rather than the same one. It costs far more to run, which is why it sits at the top of the tier list, or on its own meter. Watch the character while neither of you is speaking and you will know which kind you have: a puppet idles tidily and repeats, a generated face never quite settles.
The face is not listening while you speak. It is drawn afterwards, from a reply that is already written.
Why it stops reading as a face
Most people have a moment in the first session where the picture flips from impressive to slightly wrong. That is not fussiness. A still face has to be accurate; a moving one has to be accurate and also move the way faces move, and the mismatch between a convincing likeness and not-quite-convincing motion is the territory the uncanny valley describes. Three things tend to trigger it: the framing never changes, so the character sits at the same distance in the same light throughout; the eyes point at the camera permanently, because nothing in the pipeline knows where else to look; and the mouth is accurate while the rest of the face stays placid, since the audio told it about speech, not about meaning.
How much that bothers you is personal, which is the best reason to spend your first minutes of video on a free tier rather than on a plan bought for it. Some stop noticing by the third call; others would rather have the voice with the screen off.
Where an AI girlfriend app video call sits in the pricing
A text reply costs an app one pass through a model, a few words long. A second of video is many pictures, animated or generated and streamed to you, and that cost scales with duration rather than with what was said. A long silence in a video call costs nearly what a long sentence does.
Which is why video is almost never sold the way text is. The shapes you will meet: call minutes included per month on a tier, with packs of extra minutes sold separately; a coin balance spent per minute and topped up when it runs out, the pattern our piece on coin shops and what they really charge covers; or the feature gated to the highest tier with a soft cap nobody documents. The first is predictable. The second is where a fixed subscription becomes a variable bill.
So the number of minutes is the price, and the arithmetic to do is against your own habits rather than the headline figure. Price a normal week — two short calls, or most evenings for twenty minutes — on each tier with top-ups included. An allowance that sounds generous monthly can be two evenings. Check too whether minutes run from when the call connects or from when the character speaks, since a call left open is metered either way. On free plans, expect a demonstration rather than a feature.
Your own camera, and what it is for
The input to all of this is your microphone. The face is drawn from the audio of its own reply, so nothing needs to see you, and most apps offering the feature never ask for the camera. If one does ask, treat it as a separate decision from the call: grant it while using the app rather than always, and check what the policy says about video and images rather than assuming the paragraph on chat text covers them.
When to turn it on, and when to leave it off
Video is good at a narrower set of things than its place on the pricing page suggests. It makes a short call feel like an occasion, and if you find voice flat because there is nothing to look at, it fills that in.
It is poor at what most people actually use these apps for. It adds nothing to memory, nothing to the writing, and nothing to how well a character holds a personality over weeks, which is where the quality of a companion app lives. It also pulls you towards short sessions, since you have to be somewhere you can watch a screen and talk out loud, and it meters them while text sits under a flat fee. Get the conversation right first and treat video as something you switch on occasionally. If you are picking where to try it, the app we currently recommend is a reasonable starting point, and the test worth running is the dull one: have a short call, then go back to text for a week, and see which you missed.
Frequently asked questions
Can you video call an AI girlfriend app?
On a growing number of apps, yes, usually on a paid tier and with a limited number of minutes. What you get is a generated face animated from the audio of the character's reply — not a transmission from anywhere, and it cannot react to you while you speak.
Why are AI girlfriend app video calls sold by the minute?
Because the cost sits in the duration rather than in what was said. Every second on screen is a stream of pictures to animate or generate, so a quiet minute costs roughly what a talkative one does — hence minute allowances and coin balances instead of a flat fee.
Does the AI see you during a video call?
Not as part of how the feature works. The pipeline runs on your microphone audio, and the face is drawn from the reply it just produced. If an app asks for camera access anyway, treat that as a separate decision.
Disclosure. SweetSoul AI 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.